JMBC2010

Lecture notes of the
J. M. Burgerscentrum
Course on Combustion
September 28 October 1, 2010

Eindhoven University of Technology
Department of Mechanical Engineering
Organized by:
prof.dr. L.P.H. de Goey & dr.ir. J.A. van Oijen
prof.dr.ir. D.J.E.M. Roekaerts

Delft University of Technology
Edited by:
dr.ir. J.A. van Oijen
The cover shows a photograph of the application of PIV in a high speed laminar methane/oxygen
ame with co-ow (courtesy of K.R.A.M. Schreel).
Contents
List of Lecturers
Part 1
Introduction to Combustion L.P.H. de Goey

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Premixed ames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Non-premixed ames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Part 2
2
Introduction
Laminar Combustion
3
3
3
7
11
Governing equations L.M.T. Somers

2.1 Conservation equations . . . . . . . . . .
2.2 Constitutive relations . . . . . . . . . . .
2.3 Chemical aspects in combustion . . . . .
2.4 An example: methane mechanism, skeletal
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
13
13
15
21
26
Premixed laminar ames L.P.H. de Goey

3.1 Introduction . . . . . . . . . . . . . . .
3.2 Analysis of premixed ames . . . . . .
3.3 Flame stretch . . . . . . . . . . . . . .
3.4 Flame stability . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
33
33
33
41
48
Numerical Modelling of Laminar Flames J.H.M. ten Thije Boonkkamp

4.1 A mathematical model for laminar ames . . . . . . . . . . . . . . .
4.2 Space discretization . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Time integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4 A pressure correction method . . . . . . . . . . . . . . . . . . . . . .
4.5 Solution methods for algebraic systems . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
57
57
59
64
66
67
.
.
.
.
73
73
73
74
84
.
.
.
.
Reduced chemical models L.M.T. Somers

5.1 Introduction . . . . . . . . . . . . . . . . . .
5.2 Global reaction models . . . . . . . . . . . .
5.3 Systematic reduction technique . . . . . . . .
5.4 Mathematical systematic reduction techniques
i
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
ii
6
Contents
Flamelet-Generated Manifolds J.A. van Oijen
6.1 Introduction . . . . . . . . . . . . . . . . .
6.2 Flamelet equations . . . . . . . . . . . . .
6.3 Manifold construction . . . . . . . . . . . .
6.4 Application of FGM . . . . . . . . . . . . .
6.5 Test results . . . . . . . . . . . . . . . . .
6.6 Conclusions . . . . . . . . . . . . . . . . .
Part 3
7
10
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Turbulent Combustion
115
Introduction to turbulent combustion D.J.E.M. Roekaerts

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . .
7.2 Computational approaches . . . . . . . . . . . . . . . .
7.3 Instantaneous conservation equations . . . . . . . . . . .
7.4 Basic denitions from statistics . . . . . . . . . . . . . .
7.5 Averaged equations in RANS . . . . . . . . . . . . . . .
7.6 LES equations . . . . . . . . . . . . . . . . . . . . . . .
7.7 Closure of the chemical source term in RANS . . . . . .
7.8 Closure of the chemical source term in LES . . . . . . .
7.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
117
117
118
120
120
123
123
125
131
134
Non-premixed amelet methods L.M.T. Somers

8.1 Flamelets: introduction . . . . . . . . . . . .
8.2 Flamelet equations: derivation . . . . . . . .
8.3 Laminar counterow diusion ame(lets) . .
8.4 Presumed pdf modelling . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
139
139
139
141
146
Premixed amelet methods R.J.M. Bastiaans

9.1 Introduction . . . . . . . . . . . . . . . . .
9.2 DNS with amelet model . . . . . . . . . .
9.3 LES modeling . . . . . . . . . . . . . . . .
9.4 Preferential diusion eects . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
153
153
155
158
163
Probability Density Function (PDF) methods D.J.E.M. Roekaerts

10.1 PDF transport equation . . . . . . . . . . . . . . . . . . . . . .
10.2 Monte Carlo solution method . . . . . . . . . . . . . . . . . . .
10.3 Hybrid ow eld model . . . . . . . . . . . . . . . . . . . . . .
10.4 Modeling of scalar micro-mixing . . . . . . . . . . . . . . . . .
10.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
169
169
170
172
176
180
Part 4
11
.
.
.
.
.
.
89
89
90
94
98
103
110
Laser Diagnostics
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
185
An introduction to laser diagnostics N.J. Dam

187
11.1 Basic molecular spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
11.2 Basic laser-diagnostic techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
Contents
iii
12
Recent developments in laser diagnostics M. Ald n

e
247
12.1 Introduction and background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
12.2 Further developments and applications of planar laser-induced uorescence . . . . 249
12.3 Thermographic phosphors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
13
LDA and PIV M.J. Tummers en L.F.G. Geers

267
13.1 Laser Doppler Anemometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
13.2 Particle Image Velocimetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
14
Thermometry by coherent anti-Stokes Raman spectroscopy E.H. van Veen

14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.2 Spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.3 Thermometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.4 Experimental setups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.5 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
Application of laser diagnostics Th.H. van der Meer

15.1 Introduction . . . . . . . . . . . . . . . . . . . . .
15.2 The Delft piloted diusion ame burner . . . . . .
15.3 Experimental results . . . . . . . . . . . . . . . . .
15.4 Conclusions . . . . . . . . . . . . . . . . . . . . .
Part 5
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
287
287
287
290
291
292
.
.
.
.
295
295
295
296
308
Clean Combustion Concepts
311
16
Preheated dilute combustion D.J.E.M. Roekaerts and M.J. Tummers
17
Clean combustion concepts in furnaces E.-S. Cho, B. Danon and W. de Jong

17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17.2 MEEC furnace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17.3 Experimental results multi-burner furnace . . . . . . . . . . . . . . . . . .
17.4 Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
19
New combustion concepts in Diesel engines L.M.T. Somers

18.1 Introduction to Diesel engines . . . . . . . . . . . . . . .
18.2 Combustion modes in modern Diesel engines . . . . . . .
18.3 PCCI/LTC studies at the TU/e: Experimental research . . .
18.4 PCCI/LTC studies at the TU/e: Numerical research . . . .
Spray combustion in engines C.C.M. Luijten
19.1 Introduction . . . . . . . . . . . . . . . . .
19.2 Spray phenomenology . . . . . . . . . . .
19.3 Classication of spray models . . . . . . .
19.4 Phenomenological spray models . . . . . .
19.5 Recent developments and Outlook . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
313
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
315
315
316
320
324
.
.
.
.
333
333
334
336
340
.
.
.
.
.
353
353
358
371
373
384
iv
Contents
Part 6
Applications
391
20
Thermo-acoustics in laminar ames V.N. Kornilov
393
21
Thermo-acoustics in turbulent ames J.B.W. Kok
395
22
Gas turbines J.B.W. Kok
397
23
Fire B. Merci
23.1 Introduction .
23.2 Fire triangle .
23.3 Fire plumes .
23.4 Enclosure re
23.5 Flame spread
24
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Radiative heat transfer P.J. Coelho
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
399
399
399
400
402
405
411
List of Lecturers
Prof.dr. M. Ald n (Marcus) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lund Institute of Technology

e
Combustion Physics
P.O. Box 118
S-221 00 Lund, Sweden
+46-46-2227657
marcus.alden@forbrf.lth.se
Dr.ir. R.J.M. Bastiaans (Rob) . . . . . . . . . . . . . . . . . . . . . . . . . . . . Eindhoven University of Technology
Section Combustion Technology
P. O. Box 513
5600 MB Eindhoven
+31-40-2474836
r.j.m.bastiaans@tue.nl
Dr. P.J. Coelho (Pedro) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Instituto Superior T cnico
e
Av. Rovisco Pais
1049-001 Lisboa, Portugal
+351-218418194
pedro.coelho@ist.utl.pt
Dr. N.J. Dam (Nico) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Eindhoven University of Technology
P. O. Box 513
5600 MB Eindhoven
+31-40-2472117
n.j.dam@tue.nl
Prof.dr. L.P.H. de Goey (Philip) . . . . . . . . . . . . . . . . . . . . . . . . . . Eindhoven University of Technology
P. O. Box 513
5600 MB Eindhoven
+31-40-2472938
l.p.h.d.goey@tue.nl
v
vi
List of Lecturers
Dr.ir. W. de Jong (Wiebren) . . . . . . . . . . . . . . . . Delft University of Technology

Faculty of Design, Engineering and Production
Section Thermal Power Engineering
Mekelweg 2
2628 CD Delft
+31-15-2789476
wiebren.dejong@tudelft.nl
Dr.ir. J.B.W. Kok (Jim) . . . . . . . . . . . . . . . . . . . . University of Twente
P.O. Box 217
7500 AE Enschede
+31-53-4892582
j.b.w.kok@utwente.nl
Dr. V.N. Kornilov (Viktor) . . . . . . . . . . . . . . . . . Eindhoven University of Technology
P. O. Box 513
5600 MB Eindhoven
+31-40-2473819
v.n.kornilov@tue.nl
Dr.ir. C.C.M. Luijten (Carlo) . . . . . . . . . . . . . . . Eindhoven University of Technology
P. O. Box 513
5600 MB Eindhoven
+31-40-2475347
c.c.m.luijten@tue.nl
Prof.dr.ir. Th.M. van der Meer (Theo) . . . . . . . University of Twente
P.O. Box 217
7500 AE Enschede
+31-53-4892530
t.h.vandermeer@utwente.nl
Prof.dr.ir. B. Merci (Bart) . . . . . . . . . . . . . . . . . . Ghent University
Department of Flow, Heat and Combustion Mechanics
Faculty of Engineering
Sint-Pieternieuwstraat 41
B-9000 Gent, Belgium
+32-9-2643575
bart.merci@ugent.be
List of Lecturers
vii
Dr.ir. J.A. van Oijen (Jeroen) . . . . . . . . . . . . . . . . . . Eindhoven University of Technology

P. O. Box 513
5600 MB Eindhoven
+31-40-2473133
j.a.v.oijen@tue.nl
Prof.dr.ir. D.J.E.M. Roekaerts (Dirk) . . . . . . . . . . . Delft University of Technology
Department of Multi-Scale Physics
Reactive Flows & Explosions
P.O. Box 5046
2600 GA Delft
+31-15-2781448
d.j.e.m.roekaerts@tnw.tudelft.nl
Dr.ir. L.M.T. Somers (Bart) . . . . . . . . . . . . . . . . . . . Eindhoven University of Technology
P. O. Box 513
5600 MB Eindhoven
+31-40-2472107
l.m.t.somers@tue.nl
Dr.ir. J.H.M. ten Thije Boonkkamp (Jan) . . . . . . . Eindhoven University of Technology
Department of Mathematics and Computer Science
P. O. Box 513
5600 MB Eindhoven
+31-40-2474123
j.h.m.tenthijeboonkkamp@tue.nl
Dr. M.J. Tummers (Mark) . . . . . . . . . . . . . . . . . . . . . Delft University of Technology
Lorentzweg 1
2628 CJ Delft
+31-15-2782477
m.j.tummers@tudelft.nl
Dr. E.H. van Veen (Eric) . . . . . . . . . . . . . . . . . . . . . . Delft University of Technology
Lorentzweg 1
2628 CJ Delft
+31-15-2786925
e.h.vanveen@tudelft.nl
viii
List of Lecturers
Part 1
Introduction
Chapter 1
Introduction to Combustion
L.P.H. de Goey
1.1
Introduction
Combustion may take place in many dierent forms and circumstances. Depending on the ow type
and the type of premixing, it is possible to distinguish dierent kinds of ame structures, each with its
own characteristic features. The principle description of all gaseous combustion processes, however,
is based on the same equations: the conservation equations for mass, species mass fractions, energy
and momentum. There are two main categories of gaseous ames, which are categorised according to
the type of mixing: premixed and non-premixed ames.
Premixed ames will be considered in section 1.2. The Borghi diagram for visualising the dierent
combustion regimes in premixed turbulent combustion will be explained in section 1.2.1. Main focus
will be on the dierent laminar amelet regimes, where the ame time/length scales are so small that
the ames behave as laminar ames, i.e. as thin boundary layers embedded in the ow. The ame
front structure of the premixed ames in the laminar amelet regime can be considered as distorted
1D ame structures. The structure and burning velocity of idealised 1D ame fronts is studied in
section 1.2.2. A similar procedure is followed for non-premixed ames in section 1.3.
1.2
1.2.1
Premixed ames
Regime diagram for premixed combustion
The nature of a combustion process heavily depends on the length and time scales in the ame front
and the (turbulent) ow eld. To classify the dierent combustion regimes in premixed ames one
frequently uses the Borghi diagram, which has been modied by Peters [3] recently (see gure 1.1). In
this diagram, combustion regimes are dened in terms of the ratio of velocity scales v /sL and length
scales t / F . If the uctuations in the ow eld v are small compared to the adiabatic burning velocity
sL and if the length scales of the largest ow structures t are small compared to the ame thickness
F , the combustion takes place in the laminar ames region. The wrinkled amelet region is entered
if v /sL < 1 and t / F > 1: a laminar ame front is more or less undisturbed and wrinkled by the
turbulent ow eld. Distortions are damped by the propagation of the ame. However, if v /sL > 1
the burning velocity is not large enough to smoothen the uctuations and we enter the corrugated
amelet regime. If v /sL increases further we enter the thin reaction zones regime, where the smallest
ow distortions become smaller than the ame thickness. The boundary between the last two regimes
is given by the relation F = , where is the length scale of the Kolomogorov eddies, the smallest
eddies in the ow. A laminar premixed ame front has a ame thickness given by F = D/sL [4],
3
Chapter 1 L.P.H. de Goey
Figure 1.1: The modied Borghi-diagram for premixed turbulent combustion.
i.e. it is dictated by diusive processes in the structure, D being a typical diusion constant. The
reaction layer, where the combustion actually takes place, has a thickness which is an order of
magnitude smaller than F . As long as the smallest ow structures are larger, they cannot penetrate
the reaction layer and this remains more or less undisturbed (thin reaction zones regime). When these
ow structures become even smaller, combustion takes place in broken reaction zones: the system is
extensively disturbed, so that the reaction may take place in regions instead of in layers.
Most of the advances in our understanding of ame structure have been achieved in the wrinkled
and corrugated amelet regimes, in which the characteristic time and length scales of the external perturbations (such as turbulence) of the ame front are large compared to those governing the ame front
itself. This is no wonder, because the high temperatures encountered ( 2000 K) accelerate the chemical reactions, so that the ame fronts are very thin structures. This means that the internal structure
of the ame fronts is only weakly distorted by the perturbations in the amelet combustion regimes.
Following these ideas, a number of theoretical and numerical models has been developed recently for
combustion processes in this regime. However, as the levels of turbulence increase and the typical
ame temperatures decrease, the perturbing ow structures become comparable to the ame thickness (and vice versa), and other combustion regimes can be entered (e.g. thin reaction-zones regime
and distributed reaction zones regime). These regimes have been relatively unexplored to date. Not
only the interaction of the ame and the external distortions on a microscopic scale (turbulent eddies,
acoustic waves, etc.) behaves completely dierent, possibly leading to more frequent extinction and
ignition phenomena, also the meso- and macroscopic description drastically alters. Recent experimental evidence has indicated that, under the right conditions, turbulence and mixing processes can be so
intense that a ame front is no longer observed, implying homogeneous ignition/oxidation; phenomena of this kind have been reported in the literature as ameless oxidation or homogeneous-charge
compression ignition. These new types of combustion phenomena are of great potential practical
5
3
0.2
0.01
x 10
H2O
0.18
CH
0.009
O2
3.5
Tmax=2230 K
0.16
OH
0.008
3
0.14
CH3O 500
Xi []
CH4
0.005
CO2
0.08
0.002
0.02
HCO 20
0.003
0.04
0.004
0.06
CH2O
2.5
0.006
Xi []
Xi []
0.12
0.1
0.007
Stoichiometrisch CH4/lucht
T =300 [K], p = 1 bar
0.001
1.5
0
0.1
0.1
0.2
0.3
0.4
0.5
0.6
0
0.1
0.5
0.1
0.2
0.3
0.4
x [cm]
x [cm]
0.5
0.6
0
0.05
0.05
0.1
x [cm]
Figure 1.2: The internal structure of a 1D adiabatic stoichiometric premixed methane-air ame at ambient
conditions; (a) temperature and major species mole fractions; (b) O, OH and H and (c) CH3 , CH3 O, CH2 O and
HCO mole fractions
benet; these techniques seem to oer improved eciency and noise characteristics coupled with signicantly lower pollutant emissions. We will not consider these regime here, but we will focus on the
laminar amelet regimes in which the ame fronts are thin and only weakly internally distorted by
the ow.
1.2.2
Flame fronts and burning velocity
In the amelet regimes, the local propagation velocity of the premixed ame in a uid ow with local
velocity v is given by vf = drf /dt = v + sL n, where sL is the local burning velocity and n the local unit
vector normal to the ame front in the direction of the unburnt gases. The burning velocity sL and the
internal structure of such a ame is almost the same as the adiabatic burning velocity s0 and internal
L
structure of a perfectly at (one-dimensional) ame. These adiabatic 1D ames are considered here.
The internal structure of a at stoichiometric methane-air ame at ambient conditions is presented in
gure 1.2. The temperature prole and major species mole fractions are shown in gure 1.2a, the O,
OH and H radical mole fractions in gure 1.2b and the intermediate species CH3 , CH3 O, CH2 O and
HCO in gure 1.2c.
The adiabatic burning velocity is a function of the conditions of the unburnt mixture
s0 = s0 (pu , T u , ),
L
L
pu , T u and being the pressure, temperature and equivalence ratio of the unburnt mixture, respectively.
The physical background explaining the propagation velocity of a premixed ame depends on two
major factors: rst of all, the unburnt mixture just ahead of the ame front starts reacting only if the gas
temperature has increased up to a value very close to the adiabatic burning velocity. This observation
is related to the reaction time scale tR (1/Au ) exp(Ea /RT ) (of a second-order global reaction) which
is very large unless the temperature T almost equals T b , due to the large activation energy Ea of the
most important reactions. The exponential dependence of tR is related to the Boltzmann translational
energy distribution in the gas, indicating that only collisions with kinetic energy larger than Ea can
undergo chemical reaction. The chemical reaction time scale is also inversely proportional to the
frequency factor A of the reaction and the density of the mixture u . The second important factor
which governs the propagation of a premixed ame is the rate of preheating of the cold layer in front
of the reacting gas up to the temperature close to T b . This layer ahead of the front is referred to as
the preheating zone and is heated by molecular diusion from the hot ame region, described by the

40
35
30
SL
25
20
15
10
5
0
0.5
0.6
0.7
0.8
0.9
1.1
1.2
1.3
1.4
1.5
Figure 1.3: The adiabatic burning velocity s0 of a methane-air ame as a function of fuel equivalence ratio at
L
ambient conditions
diusivity D. As the burning velocity depends on both parameters tR and D, simple scaling indicates
that the following relation holds for s0 [1]:
L
s0 =
L
D
=
tR
u AD exp(Ea /RT b ).
(1.1)
This equation is valid for second-order chemistry and its basic form can be derived also from rst
principles (i.e. by solving the 1D combustion equations). This formula indicates the importance of
two major eects on s0 . First of all, s0 is very sensitive to changes in the (adiabatic) burning velocity
L
L
T b , which can be inuenced by changes in the unburnt gas temperature T u of the mixture (it is found
that s0 = 8 + 26.5(T u /298)2.11 cm/s for a stoichiometric methane-air ame), but also by the heat
L
loss from the ame (e.g. to a burner) and by changing the equivalence ratio. The adiabatic burning
velocity of a methane/air ame at ambient conditions is presented as function of the equivalence ratio
in gure 1.3. The burning velocity rapidly decreases if the equivalence ratio increases above 1.05
or decreases below 1.05, due to changes in T b . A second important inuence on s0 follows from
L
Eq.(1.1): the pressure dependence. Note that u pu and D 1/pu , so that Eq.(1.1) indicates
that s0 is independent of pu . This can be shown rigorously as well for a pure set of second-order
L
reactions. In reality it is found, however, that s0 decreases with pu roughly as s0 p0.5 for methaneu
L
L
air. This decreasing burning velocity as the pressure increases is caused by the increasing importance
of third-order (mainly chain-breaking) reactions which enhance the depletion of the radical pool and
thus cause a slowing reaction process. Note that although s0 decreases with pu , the mass burning rate
L
m0 = u s0 increases proportional with p0.5 .
u
L
Figure 1.4: Regimes in non-premixed turbulent combustion.
1.3
1.3.1
Non-premixed ames
Regime diagram for non-premixed combustion
Non-premixed ames are also sometimes referred to as diusion ames, because diusive mixing
mostly is the rate-controlling process, due to the high temperatures encountered in these ames.
Again, as in the premixed laminar amelet regime, this indicates that chemical reaction times are
very short due to the high temperatures. However, it depends on the situation whether this is true:
other regimes are also possible, although the majority of combustion applications takes place in the
amelet regime. A regime diagram for combustion in non-premixed gases is introduced by Peters [2]
and presented in gure 1.4. In premixed systems, the regime is constructed by considering the ratio of
velocity scales and length scales of chemistry and turbulence. The regime diagram for non-premixed
combustion is completely dierent than the one for premixed systems, because non-premixed ames
do not have a characteristic velocity scale (i.e. burning velocity) and not a characteristic length scale,
so that only time scales can be compared directly. As a measure for this, we introduce the Damk hler
o
number Da = t /tR , where t is the typical time scale of the smallest ow scales and tR the chemical
time scale. If the characteristic time of the smallest ow structures (e.g. Kolmogorov eddies) is larger
than the chemical time scale, we have thin fronts embedded in the ow and we refer to this regime
as the laminar amelet regime. The characteristic chemistry time scale tR is dened here as the point
where the ow time scales are so small that the ame extinguishes. This automatically means that
ame extinction occurs if Da < 1 and that Da > 1 refers to the amelet regime. The thickness of the
amelet is directly related to the time scale and is not an independent variable. However, it is possible
to consider a typical variation of the degree of mixing of fuel and oxidiser Zst at the ame front as an
independent variable. If the local uctuation in the degree of mixing is large compared to Z (indicating the range in mixing where chemistry is important), the ame fronts are thin and remain separated
when moved around by the turbulent structures. On the other hand, if this variation in mixing is small
the gases become (almost) premixed, the front thickness is large and turbulent uctuations lead to
connecting ame fronts. The major characteristics of the thin non-premixed amelet structures are
treated in the next sub-section.
Z=c1
Yfu
1.3.2
Z=c2
Z=Zst
Z=c3
Figure 1.5: Mixing of two streams with fuel and oxidiser
Yox
Structure of non-premixed ames
The position of a thin non-premixed ame in a ow depends on the mixing of the fuel and oxidiser
gases. Let us rst consider a pure mixing process of fuel and oxidiser without combustion and suppose
that YF and Yox are the local mass fractions of fuel and oxidiser gases in the system. The pure mixing
of fuel and oxidiser issuing from two separate streams 1 and 2 (see gure 1.5), then can be described
by the so-called mixture fraction Z, dened as
Z=
sYF Yox + Yox,2

,
sYF,1 + Yox,2
(1.2)
with YF,1 the fuel mass fraction in stream 1 and Yox,2 the oxidiser mass fraction in stream 2. Furthermore, s is the so-called stoichiometric coecient, indicating the ratio of oxidizer mass and fuel mass
which would be necessary for complete combustion. Note that Z does not change due to chemical reaction, because if YF decreases by an amount x due to chemical reaction, Yox decreases by an amount
sx, so that sYF Yox and thus also Z remains constant. Z in Eq.(1.2) is normalised in such a way that
Z is equal to 1 in stream 1 and 0 in stream 2.
The mixing of fuel and oxidiser can be predicted by solving the transport equation for Z, which
describes pure mixing of the gases from the two streams without combustion. It is interesting to note
that chemical reaction does not inuence Z, so that the Z-eld does not change if the mixing gases
also react. The iso-plane Zst = Yox,2 /(sYF,1 + Yox,2 ), i.e. where YF = sYox , is an important iso-plane in
the mixing ow because this is the plane where the fuel and oxidizer would burn optimally if the ow
would be reacting, where the highest temperature would be found and where the chemical reaction
time would be minimal. So this is the plane which the ame front nally tries to reach after ignition
of the mixing ow.
In case of an innitely fast reaction, the stoichiometric iso-plane Zst of Z gives rise to a completely
burnt mixture with YF,b = Yox,b = 0 after ignition. Away from this plane, YF,b or Yox,b becomes zero,
so that the reaction layer becomes innitely thin in this case. Eq.(1.2) indicates that there is a linear
relation between the mass fractions YF,u and Yox,u of the non-reacting case on the one hand and Z on
the other. This relation is presented in gure 1.6a. Furthermore, there also exists a linear relation
between the species mass fraction YF,b , Yox,b , the temperature and the product species mass fractions
for the reacting case on the one hand and Z on the other hand in case of an innitely fast combustion
sYF,b +Yox,2
reaction. This can be understood if one realises that Yox,b = 0 if sYF,u > Yox,u leading to Z = sYF,1 +Yox,2 ,
Y
+Y
ox,2
while YF,b = 0 if sYF,u < Yox,u , giving Z = sYox,b+Yox,2 . Similar relations can be derived for the product
F,1
species mass fractions and the temperature as function of Z. Some of these relations are depicted in
gure 1.6b. From this gure, it is possible to derive the mixture composition and temperature in the
ame if the Z-eld is given or known.
0.4
0.4
YOx,b
Z=Z
st
0.35
Z=Zst
0.35
YOx,u
0.3
0.3
0.25
0.25
Yi []
Yi []
YCO
0.2
,b
0.2
YH
0.15
O,b
0.15
YFu,u
YFu,b
0.1
0.1
0.05
0.05
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.1
0.2
0.3
0.4
0.5
Z []
0.6
0.7
0.8
0.9
Z []
Figure 1.6: Fuel and oxidiser mass fractions as a function of the mixture fraction Z (a) for the pure mixing case
(YF,u and Yox,u ) and (b) for the reacting case (YF,b , Yox,b , YH2 O,b and YCO2 ,b ).
0.45
0.45
O2
T
0.4
Z=Zst
0.4
O2
0.35
0.35
0.3
0.3
0.25
Xi []
Xi []
0.25
CH4
0.2
CH4
0.2
H2O
H2O
0.15
0.15
0.1
0.1
0.05
0
0.2
CO2
CO2
0.05
0.15
0.1
0.05
x [cm]
0.05
0.1
0.15
0.2
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Z []
Figure 1.7: Internal structure of a methane-air diusion ame in a stagnation ow (a) as a function of distance
x and (b) as a function of mixture fraction Z.
In case of a nite reaction time, the ame front around the stoichiometric iso-plane Zst obtains a
nite thickness, since YF,b and Yox,b are not completely zero at Z = Zst . Figure 1.7a shows the structure
of the most important species in a non-premixed methane-air ame as a function of the position x in
a stagnation ow. The burner used for this system consists of two opposed jets with air and methane
generating a stagnation ow with a stagnation plane in the centre. The position x = 0 is chosen at
the stagnation plane. Note that Zst = 0.5 for this special case and the planar non-premixed ame then
stabilises at the position where Z = Zst . Figure 1.7b shows the structure of the most important species
in the ame as a function of Z. Note the striking correspondence with gure 1.6b.
References
[1] de Goey, L.P.H., Somers, L.M.T and Bongers, H., Verbrandingtechnologie, TU/e course (2002).
[2] Peters, N., ERCOFTAC summerschool on Laminar and Turbulent Combustion, 1992.
10
[3] Peters, N., Turbulent Combustion, Cambridge University Press, Cambridge, 2000.
[4] Warnatz, Maas and Dibble, Combustion, Springer Verlag Berlin, 1996.
Part 2
Laminar Combustion
11
Chapter 2
Governing equations
L.M.T. Somers
2.1
Conservation equations
Flames and other combustion processes fall in the area of chemically reacting ows, which are governed by a set of conservation equations describing the ow, the chemical species mass fractions and
the enthalpy. The momentum and continuity equations are given by
(v)
+
t
(vv) +
+
t
() = p,
(v) = 0,
(2.1)
(2.2)
, p, and v being the mass density, pressure and the ow velocity and being the stress tensor. The
conservation equations for the N species mass fractions Yi are
(Yi )
+
t
(vYi ) +
(Vi Yi ) = i , (i = 1, ..., N).
(2.3)
In this equation Vi is the diusion velocity of species i, which generally is related to the local gradients
of all species mass fractions Yi , the temperature T and the pressure p described by the Stefan-Maxwell
equations. Finally, i are the chemical source terms, with contributions from all reactions in the
mixture. The set of conservation equations is closed by the balance equation for the specic enthalpy
h, given by
(h)
p
+ (vh) = q : ( v) +
+ v p.
(2.4)
t
t
For a mixture the specic enthalpy h is dened as the sum over the specic enthalpy hi of each species
i:
N
h=
hi (T )Yi
(2.5)
i=1
with the normal denition of the specic heat of species i according to dhi = c p,i dT . Note that the
chemical source terms i do not appear here! Equivalently, one could also use the conservation of
specic internal energy e = h p/,

(e)
+
t
(ve) = q : ( v) + p ( v) ,
13
(2.6)
14
Chapter 2 L.M.T. Somers
or for that matter the conservation of specic total energy, e = e + v2 /2,
( )
e
+
t
(v ) = q
e
( v) +
(pv) + v g,
(2.7)
to complete the set of conservation equations. The mixture specic internal energy e is dened as the
sum over the specic internal energy ei of each species i:
N
e=
ei (T )Yi
(2.8)
i=1
Now the specic heat cV,i of species i is dened according to dei = cV,i dT .
The equations (2.4,2.6,2.7) are equivalent and can be transformed into each other by using the
relation between e and h and the conservation of momentum (eq. 2.1). It is merely a matter of taste. In
some situations it is more natural to use conservation of energy, for instance in constant volume systems. In most open combustion systems however the conservation of enthalpy is to be preferred. In
that case the pressure is approximately constant and the terms involving pressure in eq. 2.4 can be neglected. If also the eect of viscous forces is neglected this is referred to as the so-called Combustion
Approximation.
An often used quantity in theoretical analyses is the specic element mass fraction. An element
mass fraction Z j is dened by [24]
N
Zj = Wj
ji
i=1
Yi
, ( j = 1, . . . , Ne ),
Mi
(2.9)
where the ji -coecients denote the species composition (i.e. the number of atoms of type j in species
i) and where W j is the molar mass of element j. Ne is the number of elements in the ame. The
conservation equations for Z j read
(Z j )
+
t
(vZ j ) =
w ji (Yi Vi ) , ( j = 1, . . . , Ne ),
(2.10)
i=1
stating that element mass fractions in a ame are only inuenced by convection and diusion and not
by chemical reactions! In equation (2.10) we used the notation w ji = W j ji /Mi , denoting the mass
fraction of element j in species i. These Ne conservation equations are, of course, not independent of
the set (2.3).
The set of equations is not closed unless appropriate models for the heat-ux q, the stress tensor
, the diusive velocities Vi , an equation of state and a description of the chemical reactions as they
appear in i are introduced. This will be done in the next sections of this lecture. Only then the
equations above can be solved directly (Direct Numerical Simulation, DNS) if they are supplemented
with sucient boundary conditions and initial values. The ame fronts, however, are generally very
thin and a very ne mesh and small time steps are required to resolve all the length and time scales
in the ame front. This makes DNS of laminar ames very expensive. If the ow eld is turbulent,
additional time/length scales of the turbulent ow eld should also be resolved, which is even more
expensive if these are smaller than the ame scales. If the mesh is larger than the smallest ow scales,
it is only possible to resolve the largest turbulent structures. Such Large-Eddy Simulations (LES) are
less expensive, but one has to implement closure models for the dissipation of turbulent scales on the
nest grid level. If the ame scales are also smaller than the mesh size, one has to implement models
Governing equations
15
for the combustion in these smallest scales as well. Accurate models for this are still lacking. For
these reasons, most combustion models for turbulent ames are based on Reynolds Averaged NavierStokes (RANS) equations instead of the above equations, i.e. transport equations which are averaged
to get rid of the small scale uctuations. As a lot of unclosed terms for correlations between dierent
uctuations arise in these averaged equations, for which one has to implement closure models as well,
the system is supplemented with additional equations for the transport of these uctuations. These
models will be studied in other lectures.
Depending on the boundary conditions, dierent ame structures arise. If the gaseous fuel is
premixed with air one nds premixed ame fronts which propagate in the ow eld. Diusion ame
structures do not propagate and are completely dierent in nature. They are found when the fuel gas
is not mixed before the combustion chamber: the fuel mixes with air due to diusive processes in the
diusion ame fronts and burns immediately afterwards. Premixed and non-premixed combustion
systems will be discussed separately in the sections 1.21.3.
2.2
Constitutive relations
Concentrations of all constituents appear in the expressions for the diusive ux of heat (q), species
(V i ) and momentum (). This observation is not specic for the reactive nature in combustion processes but also holds for non-reacting mixtures. The set of equations (2.1, 2.2, 2.3) and (2.26) on
pages 13 and 18 is not closed. In the conservation equations for the four unknowns the density
, velocity vector v, specic internal energy e and the mass fractions Yi , new quantities have been
introduced. For these, being the pressure p and the stress-tensor in the Navier-Stokes and the energy equation, the heat-ux q in the energy equation and the diusion velocity V i in the convectiondiusion equation, relations have to be formulated such that the number of equations is equal to the
total number of unknowns. This closure is obtained by posing so-called constitutive relations.
In most combustion problems each species can be considered to behave as a perfect gas. The
partial pressure is then given by
pi = ni RT (= nXi RT )
with R the universal gasconstant (8.314 [J/mole/K] ) and ni the molar density [mole/m3 ] (or concentration) of species i. For later use the concentration fraction (or mole-fraction) Xi is introduced,
Xi = ni /n. It is simply related to the mass fraction, Xi = Yi M/Mi , where M is the average molar mass
and Mi the molar mass of species i. The n is the number density, or molar density of the mixture
(moles per volume).
According to Daltons law the static pressure is equal to the sum of the partial pressures thus
leading to,
p = nRT,
(2.11)
N
where the equality i=1 Xi = 1 is used. This provides a unique relation between pressure and temperature. This relation can be used to relate all remaining constitutive relations to the set (, Yi , T and v),
as is mostly done in combustion modelling.
The stress tensor of the mixture is identical with the expression for a single-component Newtonian uid, i.e.,
= [( v) + ( v)T ] + 123( v)I
(2.12)
where is the dynamic viscosity of the mixture. In fact, this expression is formally not correct since
terms involving the gradients of the diusion velocity, V i , are neglected with respect to those of the
average velocity v. Otherwise the assumption i = Yi with i the dynamic viscosity of species i,
16
has to be posed, to derive equation (2.12) rigorously [1]. As shown by Hirschfelder et al. [2] this
stress-tensor equation is the rst Chapmann-Enskog approximation to the more complete expression.
The most accurate expression for the diusion velocity, V i , appearing in the convection-diusion
equation yields [3],
DT
n
Vi = i
T+
Yi T
Xi
M j Di j d j ,
i = 1, . . . , N
(2.13)
j=1; j i
where the thermal diusion coecient DT and the generalized diusion coecients Di j are used. The
i
rst term denotes the eect on diusion caused by temperature gradients (Soret-eect). The second
term1 involves the quantity d j which takes the eect of concentration- and pressure gradients into
account, i.e,
dj =
=
1
p,
p
Mj
ln p,
Xj + Xj 1
M
Xj + Xj Yj
j = 1, . . . , N
(2.14)
j = 1, . . . , N
This expression is only valid within the perfect gas assumption mentioned earlier. More general
expressions can be found in [2, 4]. It should be noted that the pressure dependence disappears from
(2.14) if all molar masses are equal.
Due to the multi-component nature of the medium, the expression for the heat ux q becomes
fairly complex,
N
N
DT
i
q = T +
hi Yi V i RT
di .
(2.15)
Xi Mi
i=1
i=1
where hi is the specic enthalpy of the i-th component, formally given by
T
hi (T ) =
h0
i
c p,i (T )dT
(2.16)
T0
with h0 the enthalpy at the reference temperature T 0 . The specic heat capacity, c p,i (T ), of species
i
i is assumed to be a function of temperature only. Pressure dependence is usually very small and
will not be taken into account2 . The rst term in equation (2.15) is the ordinary single-component
uid heatux. It should be noted that in a mixture the conductivity, , will not be the proportionality
factor between heat conduction and the temperature gradient only. In fact, the second term, which
incorporates the enthalpy transport due to the mass diusion, also contains a term proportional to the
temperature gradient, as can be seen in the expression for the diusion velocity (2.13). The last term
in (2.15) is the direct eect of concentration- (Dufour-eect) and pressure gradients on the heat ux.
In fact the Dufour-eect in the heat ux is the counterpart of the Soret-eect in the diusive ux3 .
In principle models for all constitutive relations are now determined and the problem is closed.
Only the thermodynamical properties, e.g. hi , ei or c p,i , and transport properties, e.g. ,DT or Di j , are
i
unknown at this point. Formal expressions for these quantities can only be derived rigorously from
1
There appears to be some discussion lately on whether the

j=1; j i
2
3
should not read
instead.
j=1
Furthermore, there is very little accurate experimental data available to justify modelling it at all [5].
Heatux driven by concentration gradients vs. massux driven by temperature gradients (Onsager).
Governing equations
17
molecular theory [2, 4, 5]. Typically, these properties involve the evaluation of so-called collision
integrals which are very costly to compute and should therefore be avoided in numerical simulations
of reactive ow phenomena. Excellent treatments on the principles can be found in the latter citations
or in [3] where the expressions are extended to incorporate quantum-mechanical molecular states as
well. However, most commonly data as tabulated in thermodynamical databases provided with the
CHEMKIN-package [6, 7, 8] are used. These expressions will be treated in the section (2.2.2) of this
lecture.
2.2.1
Simplied models for the diusive ux
In the so-called Combustion Approximation the term due to pressure gradients may be neglected.
Furthermore, thermal diusion is normally very small as well and will be neglected4 . However, the
diusion velocity of a species still depends on the concentration gradients of every other. This is
especially problematic in theoretical analysis of ames, in numerical modeling this is not a priori a
problem. Within the earlier assumptions, it can be shown that the expression (2.13) can be rearranged
to yield [9, 10, 3],
N
Xi X j
Xi =
(V j V i )
(2.17)
Di j
j=1
which is commonly referred to as the Stefan-Maxwell equation. This expression is much simpler to
evaluate since it involves the binary diusion coecients, Di j , which are virtually independent of the
composition. For this reason only, the Stefan-Maxwell equation are to be preferred above the original
expression for the diusion velocity (2.13). The latter involves the generalized diusion coecients5
which depend explicitly on the composition in a complicated and CPU-intensive manner6 .
For this reason in most laminar ame studies the diusion ux is cast in a Fick-like expression,
Vi =
Dim
Xi .
Xi
(2.18)
where the so-called mixture-averaged diusion coecient, Dim , describes the diusivity of the species
in the mixture. An exact derivation of this equation does not exist. In fact, it can be derived only from
the Stefan-Maxwell equations if it is assumed that all V j ( j i) take a common value V. Substituting
this, and making use of the relation,
Yi V i + (1 Yi )V = 0
which is then true by denition, the Stefan-Maxwell equations (2.17) can be rearranged to yield,
Dim =
4
5
(1 Yi )
X j /Di j
N
j i
(2.19)
Only for very light species (H2 , H or He) this can have a considerable contribution to the total diusive ux O(20 %).
The relation between the two is known and relatively simple for a ternary mixture
D12 = D12 1 +
X3 [(M3 /M2 )D13 D12 ]

X1 D23 + X2 D13 + X3 D12
and similar relations for D21 , D23 , D32 , D13 , D31 .

6
To rst order the generalized diusion coecients involve a (N N)-matrix inversion for every species thus leading to
O(N 3 ) operations per gridpoint [11].
18
where the coecients Di j are the earlier mentioned binary diusion coecients. Strictly, in a multicomponent ow this expression only applies7 for a trace species and is therefore often referred to
as the trace-species approximation. In can be expected that this approximation applies quite well
to modelling atmospheric fuel-air combustion processes since approximately 80% Nitrogen (N2 ) is
present under these circumstances. For certain applications the species balance equation is entirely
written in terms of mass-fractions. To arrange that the expression (2.18) is rearranged according to,
Mi
D
X
im i
M
Mi
M
= Dim (Yi )
Mi
M
Yi V i =
= Dim
(2.20)
(2.21)
Yi + Yi ln( M)
(2.22)
where the identity Yi /Xi = Mi / M has been used. The last term on the right-hand side is in CH4 - air
combustion negligible since the average molar mass is almost constant.
Substitution of the trace species diusive ux in the balance equation thus gives an expression
for the species mass-fraction which is very similar to the temperature equation (equation (2.26) on
page 18),
Yi
+ v Yi (Dim Yi ) = i .
(2.23)
t
In many theoretical studies this feature is used explicitly to simplify the solution considerably. However, note that within these approximations the diusive uxes no longer sum to zero due to the
assumptions made. Thus formally, the continuity equation no longer holds. Some authors correct this
deciency by calculating the mean overall diusion velocity according to V r = Yi V i and substitute each V i by V c = V i + V r . Others replace the balance equation, (2.23), for the species which is
i
N1
most abundant (N2 in many cases) with YN2 = 1 i=1 Yi . Thus mass conservation is used explicitly
and the continuity equation is forced to hold.
Further simplication of the diusive uxes is possible if the so-called constant Lewis-number
approximation is applied. The Lewis number is dened as the ratio between enthalpy diusivity (see
eq. 2.4) /c p and species diusivity Dim
Lei =
c p Dim
(2.24)
In the late eighties and nineties this approximation was evaluated for methane-air mixtures[25, 26].
It appeared that a constant Lewis number applies very well for these ames and does not yield large
errors in ame properties. Substituting the constant Lewis number approximation in the species conservation equation now yields a very often encountered version of this equation:
Yi
+ v Yi
t
Y i ) = i .
c p Lei
(2.25)
Main application of this version of the species conservation equation is found in theoretical analyses. It can be used for instance to deduce a simpler version of the enthalpy conservation equation.
As furthermore pressure and viscous eects are neglected, the equation for h can be simplied considerably, leading to
(h)
+
t
7
(vh)
Or in a binary mixture, but that is trivial.
h =
cp
N
i=1
1
1
Lei
hi Yi .
cp
(2.26)
Governing equations
19
Furthermore, in the case of equal (e.g. unit) Lewis numbers (Lei = c or Dim = D), the conservation
equation (eq. 2.10) for element mass fractions Z j all behave the same. Even more, if the Z j are all
scaled between 0 and 1 they are described by a single transport equation for the so-called mixture
fraction Z = Z j :
(Z)
+ (vZ) = (D Z).
(2.27)
t
Normally, Z is scaled such that it is 0 in the oxidizer stream and 1 in the fuel stream.
2.2.2
Evaluating thermodynamic and transport properties
The evaluation of transport properties of any multi-component mixture is in general very complex,
and normally very CPU-intensive [11, 3]. Especially the latter argument is of major importance in
ame simulation. Therefore, many formulations exist with a dierent levels of approximation. In the
early eighties some investigators have tested these formulations [12, 13]. The expressions presented
below are a compilation of the (semi-empirical) formulae used by many authors. They are believed to
yield accurate results for the respective uxes, and do not use excessive CPU time.
For the mixture averaged viscosity, , in most standard laminar ame packages (Sandia-codes),
the semi-empirical formula of Wilke [14] is used,
N
=
k=1
k
1 + 1/Xk
N
j=1, j k
(2.28)
X j k j
where k are the single species gas viscosities. The quantity k j is given by,
Mk
1
k j = 1 +
Mj
8
1
2
1 +
1
2
Mj
Mk
1
4
The mixture averaged conductivity, , can be put in a somewhat related form [15],
N
=
k=1
1 + 1/Xk 1.065
N
j=1; j k
(X j k j )
which, however, is not used frequently in numerical combustion studies. Now the k j is given by the
expression,
1
1 2
1
2
j 2 Mj 4
Mk
1
k j = 1 +
1 +
.
Mj
k
Mk
8
In most studies however, instead a semi-empirical formulation is applied [16],
N
1
N

Xi /i
Xi i +
=
i=1
(2.29)
i=1
which involves only the pure species conductivities. According to Coee and Heimerl [12] this gives
results with errors in the range of a few percent. In simulations where viscosity is not evaluated
the latter expression is to be preferred because then the functions k j do not need to be evaluated.
This is generally true for the simulation of premixed one-dimensional ames, in multi-dimensional
20
simulations there is no preference, since then the viscosity of the mixture will have to be evaluated
anyway.
The properties, Dim and c p , are already introduced earlier in this lecture, and are presented for
matters of completeness only,
(1 Yi )
X j /Di j
Dim =
i = 1, . . . , N
N
j i
(2.30)
cp =
(2.31)
Yi c pi
i=1
and complete the mixture averaged thermodynamical and transport properties.

The pure species quantities, c pi , that arise in the averaged formula, (2.31), and, hi , in the chemical
source term of the energy equation (equation (2.26) on page 18) are well tabulated [8]. The properties
are cast in a polynomial form presented below,
c pi
=
R
hi
=
RT
en,i T n1
n=1
5
n=1
i = 1, . . . , N
(2.32)
en,i n1 e6,i
T
+
n
T
i = 1, . . . , N
with R the universal gas constant.

The binary diusion coecients can not be tabulated because diusivity is not an intrinsic property. They depend on the (Lennard-Jones) molecular potentials of the two species considered. Same
holds for the viscosities and the conductivities as well, since these are expressed in terms of binary
diusion coecients. Instead, the Lennard-Jones potential data for each species, necessary to evaluate
these properties, are tabulated in the STANJAN-tables [6]. Then for a specic problem, appropriate
tting procedures, e.g. provided with the CHEMKIN-package [6], are used to put them in a similar
polynomial form yielding,
P
ln(i ) =
an,i (ln T )n1
i = 1, . . . , N
bn,i (ln T )n1
i = 1, . . . , N
n=1
P
ln(i ) =
(2.33)
n=1
P
ln(Di j ) =
dn,i j (ln T )n1
i = 1, . . . , N, j = 1, . . . , N, j
n=1
This procedure has to be performed once for every mixture, in contrast with the coecients in (2.32)
which can be evaluated for every species once and for all.
Now, the set of equations describing reacting ows are almost in a closed form. The only terms
not specied yet, are the chemical source terms i . Topics associated with this item are treated in the
remaining part of this lecture.
Governing equations
2.3
21
Chemical aspects in combustion
In a previous part of this lecture general reactive ow equations are introduced, but the chemical source
terms appearing in the species balance equations are not specied yet. In combustion of gas, these
involve so-called chain-reactions, which means that the oxidation of the fuel is composed of many
elementary steps. These steps, or elementary reactions, have a physical relation with microscopic
processes, in contrast with overall chemical reactions where the reaction equations merely state the
overall conversion ratios.
For methane-air mixtures the elementary steps are known and tabulated. Complete CH4 chemical
models consist of approximately 36 species and 210 reactions [27]. Some general principles, present
in these detailed chemical models, will be claried in this section. These principles apply to all elds
of reactive ows.
2.3.1
General elementary reactions
In general, any elementary reaction can be written in the form,

1 A1 + 2 A2 + . . . N AN 1 A1 + 2 A2 + . . . N AN
(2.34)
where Ai represents one of the N species present in the model and i i the number of molecules
Ai converted in the elementary reaction. Since neither nuclear ssion, nor fusion processes will be
considered, each reaction conserves the total number of elements (C, H, O and N in hydrocarbon air
systems).
The formal reaction (2.34) implies that,
qj
qi
=
(i i ) ( j j )
for any pair i, j of particles in the reaction. Here, qi is dened as the net rate of increase of species i
(in moles per volume8 per second). Its now common practice to dene the reaction rate q, as being
equal to the qi for a species when (i i ) is equal to one. In return one then obtains,
q j = ( j j )q.
(2.35)
The phenomenological law of mass action now states that a reaction rate is proportional to the
product of the concentrations, ni , which leads for the formal reaction (2.34) to the expression,
N
q=k
ni i
(2.36)
i=1
where the proportionality parameter k, sometimes is referred to as the specic reaction rate constant.
The reverse reaction of (2.34) often plays an important role as well,
1 A1 + 2 A2 + . . . N AN 1 A1 + 2 A2 + . . . N AN .
Therefore, in ame modelling, it is convenient to rewrite the combination of both to,
1 A1 + 2 A2 + . . . N AN
8
1 A1 + 2 A2 + . . . N AN
The volume is not specied but in chemical literature often cm3 is used.
(2.37)
22
where the symbol
indicates the reversible nature. The net reaction rate of (2.37) is now given by,
N
q=k
ni i
i=1
ni i
(2.38)
i=1
The kf and kr are now the specic reaction rate constants of the forward and reverse reactions, respectively.
N
In a practical gas-phase combustion system the reaction order, i=1 i , of the reactions involved in
the chain mechanism will not exceed three. At low densities, or pressures < 100 bar, the probability
of a collision between four or more molecules is so small, that these higher order reactions may
be neglected. The remaining three reaction types, (rst, second and third order), are treated in the
following subsections.
2.3.2
Bimolecular reactions
The most common and least complicated type of reaction is the so-called bimolecular reaction e.g.
CH4 + H
CH3 + H2
where the methane molecule is attacked by an hydrogen radical to form a methyl radical and molecular
hydrogen and reversely. The reaction rate is now given by,
q = kf [CH4 ][H] kr [CH3 ][H2 ]
where [i] is an often used notation for the molar density ni . Normally in reaction kinetics units9 are
given in [ [moles/cm3 ] ]. The specic reaction rate constants are generally not constant. As argued
by Arrhenius (1889) an appropriate representation is given by,
k = BeEa /RT
(2.39)
where the frequency factor B and the activation energy Ea do not depend on the temperature. However,
some reactions show so-called non-Arrhenius behavior and extensive studies learn that the Arrhenius
form cannot represent this behavior correctly [17]. Therefore, usually a slightly modied form of the
reaction rate constants is chosen according to,
k = AT n eEa /RT
(2.40)
which is normally used in literature. This extra T dependence is often attributed to eects of internal
molecular orientation and internal molecular energy on the reaction probability.
2.3.3
Dissociation and recombination reactions
The two other important reaction types are the rst-order and its counterpart the third-order reaction.
The rst-order reaction10 is often referred to as a dissociation reaction, e.g.,
H2 + M H + H + M
(2.41)
9
These units are SI. Two dierents sets dened within the SI regulations referred to as MKS-system (Meter-KilogramSecond) and the in reaction kinetics more often used CGS-system (Centimeter-Gram-Second).
10
Though not truly rst order, like in particular dissociation reactions of a highly excited molecules H H + H.
2
Governing equations
23
These uni-molecular reactions are believed to follow the mechanism as proposed by Lindemann. This
mechanism, in itself a chain-reaction, proceeds via two paths, being an excitation reaction (I),
H2 + M
H + M
2
(I)
which in fact is bimolecular reaction of the hydrogen molecule with any other denoted by M, producr
f
ing an intermediate, excited H2 molecule. The specic reaction rates are k0 for the forward and k0 for
the reverse reaction (de-excitation reaction). The excited hydrogen molecule then decomposes via a
true uni-molecular reaction (II),
H H + H
(II)
2
f
with specic reaction rate k . Though the intermediate species (H ) can have dierent, energetically
2
excited states (quantum states), it is common practice to treat all with the same reaction rate [18]. The
overall reaction rate for the dissociation reaction 2.41 is now composed of two steps,
f
r
qI = k0 [H2 ][M] k0 [H ][M]
2
f
qII = k [H ].
2
(2.42)
(2.43)
The stoichiometric relations for both molecules now give,

qH2 = qI
(2.44)
qH = qI qII ,
2
(2.45)
where it should be noted that qI contains the H concentration explicitly. The application of a steady2
state relation for the excited hydrogen molecule, d[H ]/dt = 0 eliminates the excited intermediate
2
from the formal reaction rate. The dissociation rate for H2 then yields,
qH2 = kf [H2 ]
(2.46)
with a formal specic reaction rate,
kf = k0 [M]
f + kr [M]
k
0
(2.47)
independent of the excited hydrogen molecule concentration.

A similar treatment for the recombination reaction,
H + H + M H2 + M
using the elementary steps (I) and (II),
H+H
H
2
H
2
+ M H2 + M
(II)
(I)
instead, yields for the recombination reaction,

qH = kr [H][H]
with the formal specic reaction rate now equal to,
r
k0 [M]
.
r
r
k = k f
r [M]
k + k0
(2.48)
(2.49)
24
The factors in brackets in both equations (2.47) and (2.49) indicate the probability that the reactions
are completed after initiation11 . An advantage of this formulation is that this denition of kf and kr
follows the denition of chemical equilibrium,
K=
f f
[H][H] kf k k0
= r = r r.
[H2 ]
k
k k0
High and low pressure limit

The [M]-dependence of the specic reaction rates kf and kr of these types of reactions is commonly
referred to as the pressure dependence of dissociation and recombination reactions12 . Two limiting
cases can be recognized [18], one low pressure branch, [M] 0, yielding,
f
kf = k0 [M]
kr r
kr = k0 [M]
f
k
(2.50)
where the excitation reaction is thought to be rate-controlling. At high pressures, [M] , the other
limit is encountered,
kf f
kf = 0 k
r
k0
(2.51)
r
k r = k
as can be veried easily. Then the collisions between the third particle and H2 or H are so frequent
2
that the uni-molecular decomposition I and the reversible process for the recombination case become
rate-controlling. Its worth mentioning that the dissociation reaction switches from a uni-molecular
to a bi-molecular behavior going from high to low pressures, whilst the recombination switches from
tri-molecular to bi-molecular going from low to high pressures instead. Both limiting branches (2.50)
and (2.51) are plotted in gure (2.1) with the dotted lines. The relations (2.47) and (2.49) are indicated
by the top solid line.
Fall-o curve
Generally the real behavior of the specic reaction rate, solid line in gure 2.1, is not equal to the earlier derived expressions. To incorporate this experimental observed behavior a compact representation
of the fall-o curve, as it is often referred to, is mostly of practical importance. Therefore, usually
the so-called center, [M]c , of the fall-o curve located at the intersection of both limiting branches
of k is introduced. The reduced coordinate x = [M]/[M]c = k0 /k 13 oers a possibility to generate a
compact expression for the reaction rate constant,
k
x
=
k 1 + x
(2.52)
where the latter is often abbreviated by FLH (x) after Lindemann and Hinshelwood who rst introduced
this function. In fact FLH is no more than a switching function between two limiting branches of the
reaction rates indicated by a dashed curve in Fig (2.1) and equals the expression (2.47) and (2.49) as
can be veried easily.
Being that H2 + M H + M is the initiation for dissociation and H + H + M H + M for recombination
2
2
Though concentration or molar density dependence would be better.
13
Bear in mind that [M]c = k /k0 per denition
11
12
Governing equations
24
25
Chapter 2 L. M. T. Somers
k/k
FLH (x)
FLH (x) F(x)
M/Mc
Figure 2.1: Fall-o curves represented in the reduced coordinate. The dotted lines indicate the high, (k )
Figure 2.1: Fall-off curves represented in the reduced coordinate. The dotted lines indicate the high, (k )
and low pressure, (k0 [M]) limiting cases. The solid line indicated with FLH is the Lindemann-Hinshelwood
and low pressure, (k0 M) limiting cases. The solid line indicated with FLH is the Lindemann-Hinshelwood
expression. The extended expression (equation (2.53)) is shown with the second solid line indicated with
expression. The extended expression (equation (2.53)) is shown with the second solid line indicated with
FLH (x) F(x). A value of (log Fc = 0.3) (equation (2.54)) is used.
FLH (x) F(x). A value of (log Fc = 0.3) (equation (2.54)) is used.
The fall-o curve is often broader than FLH and theoretical analysis shows that this eect can be
High and low pressure limit adding a factor F(x), yielding,
accounted for reasonably by
The M-dependence of the specic reactionk rates k f and k r of these types of reactions is commonly
= FLH (x)F(x)
(2.53)
referred to as the pressure dependence of k
dissociation and recombination reactions12 . Two limiting
cases function, F(x), is in[18], one low pressuredetermined by, 0, yielding,
The can be recognized good approximation branch, M
f
f 1
log(Fc )
(2.54)
log(F(x)) k = k0 M
=
1 +r (log x)2
(2.50)
k r
k r = f k0 M
k
where log(Fc ) is the depression from the real fall-o curve at the center of the Lindemann-Hinshelwood
expression. It should be noted that more complex expressions are available that perform better in comwhere the excitation reaction is thought to be rate-controlling. At high pressures, Mand do not have a
parison with the observed behavior. They normally oer only minor modications , the other
limit is encountered,
large eect on the ame structure. An excellent survey can be found in [18, 19].
kf f
k f = 0 k
r
k0
The third party
(2.51)
r
r
So far no attention has been paid to the concentration of the inert species M, or so-called bath-gas.
k = k
It does not take part in the reaction (i i = 0) but is absolutely crucial as otherwise momentum
as can be veried easily. Then the collisions between the can, however, be any2particle present in the
and energy could not be conserved simultaneously. It third particle and H or H2 are so frequent
that the uni-molecular decompositioninert the reversible therefore, inthe recombination casegiven by
mixture. The concentration of the I and particle will process for the simplest case be become
rate-controlling. Its worth uni- and tri-molecular reactions the rate coecientfrom a uni-molecular
p/RT . However, in most mentioning that the dissociation reaction switches is dierent for every
to a bi-molecular behavior going instance, to low pressures, whilst the recombination switches from
third body. Stated explicitly, for from high
tri-molecular to bi-molecular going from low to high pressures instead. Both limiting branches (2.50)
H2 + H2 H + H + H2
and (2.51) are plotted in gure (2.1) with the dotted lines. The relations (2.47) and (2.49) are indicated
by the top solid line.
will have a higher rate than,
H2 + O2 H + H + O2 .
12
Though concentration or molar density dependence would be better.
26
On a molecular level this means that the collision probability of H2 -H2 to form an excited hydrogen
molecule (H ) is larger than of H2 -O2 for this particular reaction. Instead of introducing a specic
2
reaction rate for every collision partner, this eect is in most ame modelling studies incorporated
by introducing so-called enhanced eciencies i , which can have values ranging from 1.0 to 10.0
approximately . The concentration of the third particle M is than given by,
N
[M] =
i Xi p/RT
(2.55)
i=1
which reduces to the appropriate value p/RT when all i are equal to one. To illustrate the principles
presented here, in the next section 2.4 a specic mechanism is treated in some detail.
2.4
An example: methane mechanism, skeletal
Most hydrocarbons, in spite of the fact that the reaction paths are very dierent, show remarkable similarity in their combustion characteristics (with respect to major species proles and burning velocity).
The main reason for this is the claimed [20] dominance of the reactions,
H + O2
CO + OH
OH + O
(2.56)
CO2 + H
(2.57)
which occur in the combustion of any hydrocarbon14 . For the computation of the detailed structure of
a hydrocarbon ame a reaction mechanism covering the details of the reaction path of the alkane to
carbon-monoxide is inevitable. Also for the prediction of soot-formation and NO x , accurate proles
of the intermediates are indispensable. E.g. for the case of natural gas combustion the reaction data of
at least methane and to a lesser extent of ethane would have to be used.
In this section focus is on pure methane-air combustion for which the reaction data are relatively
well known. A set of reactions is presented that will be used also lateron in section 5. The main
aspects of this specic mechanism are treated.
For cases where the initial concentration of the fuel is not too high the path given in gure 2.2 describes the essential chain of reactions of the methane oxidation. Only under very fuel rich conditions
the path to the C3 and C4 hydrocarbons becomes more and more important15 . A good compilation of
all the reaction rate data available to date is given in [22] and in [23]. The latter authors claim to nd
good agreement for the burning velocities of most C1 and C2 fuels with a reaction scheme consisting
of 36 species and 210 reversible reactions. The most comprehensive and tested data-set for lower
hydro-carbon combustion nowadays can be found on the website of the GRI initiative [27].
Fortunately, in fuel-lean to stoichiometric conditions good agreement with experimental data can
be found if one considers the C1 -path only [25]. This leads to a large reduction in the number of
species and reactions involved which is the main reason that we consider lean methane-air ames
rst. In table 2.1, page 30, this mechanism is presented. It is sometimes referred to as a skeletal
mechanism since it represents only a minimal subset of the complete set. However, it is claimed
14
Warnatz claims that the global combustion phenomena (burning velocity, quenching rate etc...) of any alkane can
be modelled within a factor of two, by using only the H2 -O2 -CO mechanism, a realistic reaction set covering Alkane +
O, H, OH
Alkyl + OH, H2 , H2 O conversion followed by an innitely fast Alkyl to CO reaction.
15
A complication in higher hydrocarbon fuels (> C2 ), is the existence of dierent isomers of the alkyl, formed after
decomposition. Each of these isomers ( e.g. n C3 H7 versus i C3 H7 in the case of propane decomposition) has its own
formation rate.
Governing equations
Modelling Approaches to Chemical and Transport processes
CH4
C 2 H5
CH2 O
C 2 H4
HCO
CH3 O
C 2 H6
CH3
27
27
C 2 H3
CO
C3
C 2 H2
C2 H
CH2
CO
C3
CO2
Figure 2.2: Main reaction path inin methane comFigure 2.2: Main reaction path methane combustion [21] for not too fuel-rich initial conditions.
bustion [21] for not too fuel-rich initial conditions.
The dashed box describes what isis usually referred
The dashed box describes what usually referred
to as the CC-chain. The CC1 chain is shown in the
to as the 2 2 -chain. The 1 chain is shown in the
box on the left.
box on the left.
CH
CO
C1
C2
in literature that it predicts the most important ame-features well, e.g. burning velocities, species
The reaction set r1 r8 [25]. There is 25 reactions and reaction chain in the H -O2 sub-system
proles, ame temperatures in table 2.1 are in fact the main 15 species present in this 2scheme, which
which is embedded in any hydrocarbon reaction mechanism. These reaction rates are very well known
provides a considerable reduction of the problem size.
because of extensive studies in the early eighties on hydrogen ames. The reaction parameters of the
The reaction set r1 r in table 2.1 is in fact the main reaction chain in the H -O2 sub-system
chain CH4 to CO and CO 8to CO2 became well known in the late eighties when more2computing power
which is embedded in any hydrocarbon reaction mechanism. These reaction rates are very well known
and sophisticated algorithms were developed and the research focussed on more complex fuels than
because of extensive studies in the early eighties on hydrogen ames. The reaction parameters of the
hydrogen. In the following the role of several reactions in the hydrocarbon combustion process is
chain CH4 to CO and CO to CO2 became well known in the late eighties when more computing power
discussed. The terminology used can be found in most literature on combustion [22, 25].
and sophisticated algorithms were developed and the research focussed on more complex fuels than
First of the so-called initiation reactions are treated. These are the reactions that account for the
hydrogen. In the following the role of several reactions in the hydrocarbon combustion process is
ignition of the reaction chain. The only initiation reaction16 in the skeletal mechanism is a thermaldiscussed. The terminology used can be found in most literature on combustion [22, 25].
decomposition reaction (r10 ) which is considered to be the most important,
First of the so-called initiation reactions are treated. These are the reactions that account for the
ignition of the reaction chain. The only initiation reaction16 in the skeletal mechanism is a thermaldecomposition reaction (r10 ) which is considered to be theH + (M).
CH4 (+ M) CH3 + most important,
(r10 )
CH4 (+ M)
CH3 + H + (M).
(r10 )
Sensitivity analysis shows that this initiation reaction is of minor importance for concentration proSensitivity analysis shows that this initiation reaction is of minor importance for concentration proles, burning velocity and ame temperature in a steady ame. For studies where ignition is important
les, engines, engine and ame temperature is essential and For initiation reactions have to be pro(e.g. burning velocity knock, shock-tubes) itin a steady ame.otherstudies where ignition is important
(e.g. as well engine knock, shock-tubes) it is essential and other initiation reactions have to be providedengines, [22, 25].
vided as well [22, 25].
16
Another initiation reaction in CH4 -air ames, but not tabulated here, is the CH4 + O2
CH3 + HO2 reaction.
16 Another initiation reaction in CH -air ames, but not tabulated here, is the CH + O CH + HO reaction.
4
4
2
3
2
28

Once initiated the fuel breakup to carbon-monoxide proceeds via a chain of reactions (r11 to r20 ).
CH4 + H
CH3 + H2
(r11 )
CH4 + OH
CH3 + H2 O
(r12 )
CH3 + O2
CH3 O + O
(r18 )
CH3 O + H
CH2 O + H2
(r19 )
CH3 O + M
CH2 O + H + M
(r20 )
CH3 + O
CH2 O + H
(r13 )
CH2 O + H
HCO + H2
(r14 )
HCO + H2 O
(r15 )
HCO + H
CO + H2
(r16 )
HCO + M
CO + H + M
(r17 )
CH2 O + OH
This cascade of reactions is claimed to be very fast, which means that every intermediate product (i.e.
CH3 , CH3 O, CH2 O and HCO ) is rapidly consumed even at very low concentrations. The relatively
slower oxidation step of CO to CO2 proceeds via reaction r9 .
The total fuel-breakup chain in CH4 ames consumes radicals. To keep the oxidation going,
the so-called chain-branching reactions are important. These account for a fast increase of the
concentration of the radicals O, OH and H, often referred to as the radical pool. The most important
chain-branching reaction in methane-air combustion is r1 ,
H + O2
OH + O
(r1 )
which is essential in the analysis of methane-air combustion. It is called branching because out of
an initially present stable species (e.g. O2 ) and a H-radical it produces two new radicals and thereby
gives rise to an increase of the radical pool.
A third type of reactions are the shue- or chain-propagating reactions, reactions r2 r4 in
table 2.1,
O + H2
OH + H
(r2 )
OH + H2
H2 O + H
(r3 )
OH + OH
H2 O + O
(r4 )
The overall eect of these latter three reactions does not lead to an increase in the radical pool but is
merely a conversion of radicals into another. Thereby stable products (e.g. H2 O) are created.
Finally, the reaction chain terminates because of the so-called chain breaking reactions. In the
skeletal mechanism these are given by,
H + O2 + M
HO2 + M
(r5 )
OH + H + M
H2 O + M
(r24 )
H2 + M
(r25 )
H+H+M
The net eect of this reaction set reduces the radical pool because it converts the radicals into stable
products and is responsible for the formation of the nal products. Note that although reaction (r5 )
Governing equations
29
in principle is radical conserving it still is considered as chain-breaking since the created peroxideradical is less reactive than H and the rapid chain (r6 r8 ),
H + HO2
OH + OH
(r6 )
H + HO2
H2 + O2
(r7 )
H2 O + O2
(r8 )
OH + HO2
has an overall eect of a conversion of HO2 into stable products. The latter 6 reactions (r5 , r6 , r7 , r8 ,
r24 , and r25 ) convert radicals into stable species either directly (r7 , r8 , r24 and r25 ) or indirectly (r5 , r6 )
by creating radicals that open a fast chain towards stable species.
The skeletal mechanism is used in section 5 to illustrate the classical reduction technique. However, it should always be kept in mind that because of the absence of the C2 -chain, the skeletal mechanism is not considered to be adequate for rich ames. In rich ames a path to the C2 -chain through
reactions like CH3 + CH3
C2 H6 becomes more and more important, primarily due to the relatively high CH3 concentration. An illustration of the importance of these eects of incorporating the
C2 -chain in pre-mixed ame computations is presented in chapter 2 of [26].
References
[1] J.D. Buckmaster, G.S.S. Ludford, Theory of Laminar Flames, Cambridge University Press, London New York 1982
[2] Hirschfelder, J.O. Curtiss, C.F. and Bird, R.B., Molecular Theory of Gases and Liquids , JOHN
WILEY and SONS, London New York 1954
[3] G. Dixon-Lewis, Flame structure and ame reaction kinetics, II. Transport phenomena in multicomponent systems, Proc. Roy. Soc A 307, p. 111, 1968
[4] S. Chappman, T.G. Cowling, The Mathematical Theory of Non-Uniform Gases, Cambridge University Press, London New York 1970
[5] Williams, F.A., Combustion Theory, Second Edition, Addison-Wesley Publishing Company,
Redwood City 1985
[6] Kee, R.J. Dixon-Lewis, G., Warnatz, J., Coltrin, M.E. and Miller, J.A., A Fortran Computer
Code Package for the Evaluation of Gas-Phase Multicomponent Transport Properties, SANDIA
REPORT SAND86-8246, Sandia National Laboratories, Albuquerque, New Mexico 87185 and
Livermore, California 94551
[7] Kee, R.J., Rupley, F.M. and Miller, J.A., Chemkin II: A Fortran Chemical Package for the Analysis of Gas-Phase Chemical Kinetics, SANDIA REPORT SAND89-8009, Sandia National Laboratories, Albuquerque, New Mexico 87185 and Livermore, California 94551
[8] Kee, R.J., Rupley, F.M. and Miller, J.A., The Chemkin Thermodynamic Data Base, SANDIA
REPORT SAND87-8215, Sandia National Laboratories, Albuquerque, New Mexico 87185 and
Livermore, California 94551
[9] Muckenfuss, C, and Curtiss, C.F. , Thermal Conductivity of Multicomponent Mixtures, Journal
of Chemical Physics 29, p. 1273, 1958
30
Number
1
Reaction
H + O2
OH + O
O + H2
H2 + OH
H2 O + H
OH + OH
O + H2 O
5a
6
7
8
9
H + O2 + M
H + HO2
H + HO2
OH + HO2
CO + OH
A
2.000 1014
1.575 1013
1.800 1010
8.00 109
1.170 109
5.090 109
6.000 108
5.900 109
2.300 1018
1.500 1014
2.500 1013
2.000 1013
1.510 107
1.570 109
6.300 1014
5.200 1012
2.200 104
9.570 102
1.600 106
3.020 105
6.800 1013
2.500 108
3.000 109
4.000 1013
1.600 1014
7.000 1012
2.000 1013
2.400 1013
2.000 1012
1.300 1017
9.860 1014
1.000 1013
2.860 1013
2.200 1022
1.800 1018
OH + H
HO2 + M
OH + OH
H2 + O2
H2 O + O2
CO2 + H
10b
CH4 + M
CH3 + H + M
11
CH4 + H
CH3 + H2
12
CH4 + OH
13
14
15
16
17
18
19
20
21
22
CH3 + O
CH2 O + H
CH2 O + OH
HCO + H
HCO + M
CH3 + O2
CH3 O + H
CH3 O + M
HO2 + HO2
H2 O2 + M
CH2 O + H
HCO + H2
HCO + H2 O
CO + H2
CO + H + M
CH3 O + O
CH2 O + H2
CH2 O + H + M
H2 O2 + O2
OH + OH + M
23
H2 O2 + OH
H2 O + HO2
24
25
OH + H + M
H+H+M
CH3 + H2 O
H2 O + M
H2 + M
n
0.0
0.0
1.0
1.0
1.3
1.3
1.3
1.3
0.8
0.0
0.0
0.0
1.3
1.3
0.0
0.0
3.0
3.0
2.1
2.1
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
2.0
1.0
Ea
16800.0
690.0
8826.0
6760.0
3626.0
18588.0
0.0
17029.0
0.0
1004.0
700.0
1000.0
758.0
22337.0
104000.0
1310.0
8750.0
8750.0
2460.0
17422.0
0.0
3991.0
1195.0
0.0
14700.0
25652.0
0.0
28812.0
0.0
45500.0
5070.0
1800.0
32790.0
0.0
0.0
a High-pressure limit: switching function F([M]) =

1
with [M]c = 0.0063e18000.0/RT .
[M]c /[M]+1
b Enhanced third body eciencies : CH = 6.5, H O = 6.5, CO = 1.5, CO = 0.75,
i
O2 = 0.4, N2 = 0.4. Others = 1.0.
Table 2.1: Skeletal Mechanism for CH4 -air combustion. The coecients of the Arrhenius term k =
AT n exp(Ea /RT ) are given in the three columns on the right (see section 2.3.2). All units are cm, s, K and
calories according to conventions used in standard literature on combustion.
Governing equations
31
[10] Monchik, L, Munn, R.J., and Mason, E.A. , Thermal Diusion in Polyatomic Gases: A Generalized Stefan-Maxwell Equation, Journal of Chemical Physics 45, p. 3051, 1966
[11] Dixon-Lewis, G., In Combustion Chemistry (ed. W.C. Gardiner) Springer-Verlag, New York,
p. 21, 1984
[12] T.P. Coee and J.M. Heimerl, Flame structure and ame reaction kinetics, II. Transport phenomena in multicomponent systems, Combustion and Flame 43, p. 273, 1981
[13] Oran, E.S. and Boris, J.P., Detailed Modelling of Combustion Systems, Progress in Energy and
Combustion Science 7, p. 1, 1981
[14] Wilke, C.R., A Viscosity Equation for Gas Mixtures, Journal of Chemical Physics 18, p. 517,
1950
[15] Mason,E.A. and Saxena, S.C., Approximate Formula for the Thermal Conductivity of Gas Mixtures, Phys. Fluids 1, p. 361, 1958
[16] Mathur, S., Tondon, P.K. and Saxena, S.C., Thermal Conductivity of Binary, Ternary and Quaternary Mixtures of Rare gases. Mol. Phys. 12, p. 569, 1967
[17] Zellner, R., Non-Arrhenius Behavior in Bimolecular Reactions of the OH radical
The Journal of Physical Chemistry 83, p. 18, 1979
[18] Gardiner, W.C. and Troe, J., In Combustion Chemistry (ed. W.C. Gardiner, Springer-Verlag, New
York, p. 173, 1984
[19] Troe, J., Theory of Thermal Unimolecular Reactions in the Fall-o Range, Ber. Bunsenges. Phys.
Chem. 87, p. 161, 1983
[20] Warnatz, J., The Structure of Laminar Alkane-, Alkene-, and Acethylene Flames, Eighteenth
Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, p. 369, 1981
[21] Warnatz, J., The Mechanism of High Temperature Combustion of Propane and Butane, Combustion Science and Technology, 34, p. 177, 1983
[22] Warnatz, J., In Combustion Chemistry (ed. W.C. Gardiner, Springer-Verlag, New York, p. 197,
1984
[23] Egolfopoulos, F.N., Zhu, D.L. and Law, C.K., Experimental and Numerical Determination of
Laminar Flame Speeds: Mixtures of C2 -Hydrocarbons with Oxygen and Nitrogen, Twenty-Third
Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, p. 471, 1990
[24] Mauss, F.,Peters, N., in Reduced Kinetic Mechanisms for Applications in Combustion Systems
(Peters,N., Rogg,B., Eds), Lecture Notes in Physics m15, Springer-Verlag, Berlin, p. 58, 1993.
[25] Smooke, M.D. ed., Reduced Kinetic Mechanisms and Asymptotic Approximations for MethaneAir Flames, Lecture Notes in Physics, Springer-Verlag, Berlin Heidelberg, 1991
[26] Somers, L.M.T., The Simulation of Flat Flames with Detailed and Reduced Chemical Models,
Ph.D. Thesis, Eindhoven University of Technology, 1994
[27] Bowman,G., Frenklach,M., Gardiner, B., Smith,G. and Serauskas,B, GRI-Mech, Gas Research
Institute, Chicago, Illinois, USA, http://www.me.berkeley.edu/gri-mech/index.html
32
Chapter 3
Premixed laminar ames

L.P.H. de Goey
3.1
Introduction
Laminar premixed ames constitute a central theme in combustion science in at least three respects.
First, laminar premixed ames are used in many practical applications. Second, results obtained in
laminar ame theory are often used in many other combustion studies, and nally, laminar premixed
ames are very interesting from a fundamental scientic point of view. The close coupling between
transport processes and chemical kinetics leads to a ame structure in the form of a thin layer propagating through the unburnt mixture. The burning velocity of the ame front is a key parameter in
combustion. The burning velocity is very sensitive to details of kinetics and transport and is as such
used extensively to develop and validate kinetic mechanisms and combustion models. Furthermore,
laminar premixed ames display a broad spectrum of exciting phenomena, like pulsations and cellular
structures, depending on the fuel and system considered.
Recent knowledge in the eld of propagation and dynamics of premixed laminar ames is reviewed. Matalon [77] recently reviewed the current mathematical theory of ame dynamics of premixed and non-premixed ames. A physical description of the propagation of premixed ames is our
key focus, which, depending on the ow and burner system, might lead to interesting ame dynamics.
We also consider methods to control the ame propagation and dynamics, or in other words, to physically inuence the ame such that it behaves as desired. We will focus on current state-of-the-art in
modeling the physical behavior of premixed laminar ames. For that reason, we present knowledge
in the form of theoretical descriptions, but we also consider numerical results and experiments, used
to validate and conrm the knowledge obtained. Next section focusses on planar ames, while stretch
eects and instabilities are considered in subsequent sections.
3.2
Analysis of premixed ames
Current state-of-the-art knowledge of the propagation of planar deagration waves in premixed mixtures will be reviewed in this section. The ow eld considered will be a simple homogeneous plug
ow rendering a completely one-dimensional problem. Eects of distortions of the ame and the
ow in lateral directions largely inuence ame behavior, but these eects will be discussed in the
next sections. Undistorted planar premixed ames in a plug ow mostly propagate with a constant
velocity, referred to as the laminar burning velocity sL , and such ames have been studied for a long
time. The laminar burning velocity is one of the key parameters in combustion science and will be a
key parameter in this study as well. However, it is dicult to study free laminar adiabatic premixed
ames in the laboratory because such ames are hard to control. There are some methods to control
33
34
premixed planar ames, i.e., use methods to inuence the formation of premixed planar ames by
external forcing. This includes the application of well-balanced heat transfer processes, for instance
using burners and porous media. Steady propagation of planar ames is not always possible, especially in case of ames inside a closed system. This then may lead to spontaneous pulsations. These
phenomena are also discussed. This subject is closely related to oscillations in case of ames interacting with the acoustic eld. These phenomena are not discussed in this chapter but dealt with in a
later chapter.
These are the topics of this section. First, in the next subsection, we review the major structure of
a premixed planar ame in a plug ow. The main physical behavior is considered in terms of a simple
reaction kinetic mechanism. In reality, the structure is governed by detailed kinetics and transport
processes involving many species. We will only discuss the major eects of detailed kinetics and
transport on the ame structure. This is done in terms of an overall thin reaction layer combining even
thinner layers, reviewed briey in the subsequent subsection. The dynamics of planar ames stabilized on porous burners or inside porous inert media, as such inuencing the dynamics and structure
by external heat and mass transfer processes to control the ame propagation, will subsequently be
considered. Unsteady eects may arise and will be studied as well. This will be discussed in the last
subsection.
3.2.1
Structure and propagation of one-dimensional premixed ames
The propagation velocity of premixed ames has been studied since more than a century. The essential
result that the burning velocity is proportional to the square root of the reaction rate and the square
root of /c p , with the conductivity and c p the specic heat, has been recognized since rst studies
in this area, see e.g. Mallard and Le Chatelier [71]. This already indicates the importance of both
reaction kinetics as well as transport processes. The coupling between kinetics and transport leads to a
thin interior layer, separating the unburnt and burnt gases. The hydrodynamic length scales associated
with the geometry of the combustion system have a much larger length scale f . This ame front
consists of two sub-layers. In the rst, the unburnt mixture is not reacting but is heated by conduction
and this zone is referred to as the diusive (preheating) layer and has thickness T = /(c p sL ). Near
the burnt side of the ame front an even thinner reaction layer with thickness r is embedded, where
chemical reactions take place. As an example, the global structure of a stoichiometric methane-air
ame is presented in Figure. 3.1, but other hydrocarbon ames behave similarly. The separation of
scales, i.e., f
T
r , has been used to develop theoretical models; see [132] for an overview.
First theoretical models for describing the structure and propagation of laminar premixed ames have
been established almost one century ago by Zeldovich and coworkers (e.g. see [134]), but major
contributions have been developed in the period 1970-1990 by Bush et al. [19], Williams [132], Law
[67] and others, briey reviewed below. In most of these early papers, the propagation of a onedimensional adiabatic ame is studied by considering a one-step irreversible reaction and looking
for traveling wave solutions of the governing equations, i.e., f (x, t) = f (x vf t), vf = u sL for
all combustion variables f , where vf is the (constant) propagation velocity and u the ow velocity.
Introducing the coordinate x = x vf t, the transport equations for the lean reactant mass fraction Y
and the temperature T can be written as
Y
1 2 Y
= ,
x
Le c p x 2
2 T
Q
T
m
= ,
2
x
c p x
cp
(3.1a)
(3.1b)
2 PLANAR FLAMES
5
35
T
O2
Mole fractions, Temperature (10 K)
0.2
0.15
0.1
CH4
CO2
0.05
CO
0
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
x (mm)
Figure 3.1:Structure of aalean methane/air ame (equivalence ratio =0.6),=0.6), including of dierent
Figure 1: Structure of lean methane/air ame (equivalence ratio including denition denition
of differentpreheating layer T , reactionlayer rT , reaction of an inner layer with thickness and oxidation
layers: inert layers: inert preheating layer consisting layer r consisting of an inner layer with
thicknessthickness ; dotted line indicatesthickness ;prole. line indicates the heat release prole.
layer with and oxidation layer with heat release dotted
for them = sL is themass fraction Y burning temperature T can then be relative to the ame and
where lean reactant (constant) mass and the rate, i.e., it is the mass ux written as
Le = /Dc p the Lewis 2
number indicating the ratio of thermal diusivity /c p and mass diusivity
Y
D. Furthermore,1 Y = ,
m
(2.1a)
x
Le cp x2
(Y, T ) = Y AeTa /T ,
(3.2)
2T
is the m
reaction rate, Q the heat release and T a = Ea /R the activation temperature. The boundary
= ,
(2.1b)
x
cp x2
c=
conditions for (3.1) read T p T u , Y = Yu for x and T = T ad , Y = 0 for x . The
solution of (3.1) is the (constant) a translation, which means that only three boundary conditions are
where m = sLis determined up tomass burning rate, i.e., it is the mass ux relative to the ame
needed = /Dc and T and number indicating the ratio to detemine the unkown m. and mass
and Le to specify Y the Lewis the fourth condition is needed of thermal diffusivity /cpIntegrating
p
both equations
diffusivity D.over the entire domain, we conclude that the adiabatic temperature T ad is given by
Furthermore,
T ad = T u + QYu /c p . Combining the two transport equations and assuming an innitely thin reaction
layer, equivalent to T aAeTa /T , following approximation for the mass burning rate m0 can be derived
(Y, T ) = Y , the
(2.2)
[134]:
is the reaction rate, Q the heat release and T = Ea1 Ta /(2Tad )
/R the activation temperature. The boundary
a
m0 = 2 A
e
,
(3.3)
conditions for (2.1) read T = Tu , Y = Yu c p xb and T = Tad = Tu + QYu /cp , Y = 0
for Ze
for x . Combining the two transport equations and assuming an innitely thin reaction
where te subscript
layer, equivalent to 0Taindicates the following approximation for themass burning rate. For0 now,
, that this is the undistorted adiabatic mass burning rate m was
we consider adiabatic ames; while external heat losses are considered furtheron. Furthermore,
derived by Zeldovich2and Frank-Kamenetski [143]:
Ze = T a (T ad T u )/T ad is the Zeldovich number, which is a dimensionless activation temperature;

typically Ze = O(10). To derive (3.3), it was assumed that Le = 1; a detailed analysis for Le 1 and
1 Ta /(2Tad )
generalizations 2 (3.3)bfor higher order, reactions can be found in [134].
m0 = of A
e
(2.3)
cp
Ze
A dierent approach to compute the mass burning rate is based on matched asymptotic expansions
(MAE) [19]. Applying MAE, Bush and Fendell also found the expression (3.3) for the mass burning
where the subscript(in indicatesmore accurate result is due to Williams [130]. Yet another method
that this is the undistorted adiabatic mass burning rate. For

rate. A higher-order 0 1/Ze) and
now, we consider the integralames; whiledeveloped heat losses are considered furtheron. Furadiabatic analysis (IA) external by Law et al [67]. In this method the ame is
to determine m0 is
2
thermore, terms ofaa thin Tu )/Tzone thethickness number, which is a reaction layer of thickness
dened in Ze = T (Tad preheat ad is of Zeldovich and a much thinner dimensionless activation
T
temperature; typically Zeburned side of the ame. In the preheat zone advection and1; a detailed
r = T /Ze located at the = O(10). To derive (2.3), it was assumed that Le = diusion are
dominant and chemical reaction is negligible, whereas in the reaction layer diusion and reaction
balance and advection is negligible. Integrating the equations (3.1) across both layers separately and
36
approximating the integral of in a suitable way, an expression similar to (3.3) is derived. A slightly
dierent approach is presented by ten Thije Boonkkamp et al.[120]. The basic idea is to multiply
equation (3.1b) by T/x and subsequently integrate the resulting equation over the entire ame.
This way we obtain
m0 =
T ad
1
c p (T ad T u )
2Q
dT .
(3.4)
Tu
If we subsequently replace Y in by its linear interpolant LeYu /Ze and apply the Frank-Kamenetsky
transform, i.e., replace the exponent T a /T in by its rst order Taylor approximation around T ad ,
we obtain an expression similar to (3.3).
3.2.2
Multi-layer structure of one-dimensional ames
While simplicity is the elegance of the previous analysis, it carries the drawback that it is not based on
realistic chemical kinetic mechanisms. Dierences between the burning velocities in dierent mixtures are governed by dierences in kinetics and transport processes. Numerical simulation of the
one-dimensional governing transport equations is possible nowadays even for large mechanisms with
detailed kinetics and transport models. Numerical data of laminar burning velocities reproduce experimental results quite well (within roughly 10%) for lower hydrocarbon-air mixtures. This accuracy
is not surprising since sL is one of the key parameters used to t the rate parameters in the mechanisms. Furthermore, experimental data for laminar burning velocities in a large variety of mixtures
have been obtained using dierent techniques (spherically propagating ames in vessels, planar stagnation ames and planar adiabatic ames stabilized on burners). Accuracy of current experimental
data amounts to about 1 cm/s, depending on the situation. The link between measured and numerical
burning velocity data for dierent fuels and analytical results of the previous section remains obscure
due to the sensitivity to kinetics and transport processes.
Peters, Williams, Seshadri and others [94, 108, 109] have shown in the last decades that it is possible to use the theoretical framework of MAE also for ames with more detailed kinetic mechanisms.
For the case of stoichiometric methane-air ames, they started from a detailed mechanism and systematically reduced the mechanisms rst to four steps and nally to two steps. The two-step mechanism
enables the use of scale separation again by introducing two reaction layers instead of one: a very
thin inner layer with thickness where CO and H2 are produced from the fuel and a thicker oxidation
layer with thickness where CO and H2 are oxidized (see Fig.1). This involves the matching of four
zones: f
T
. The nal result for the burning velocity is given by [94]:
s2 = C
L
Tu
T0
T ad T 0
T ad T u
(3.5)
where the temperature in the inner layer T 0 is computed from the matching procedure and is given by
2
RT 0 k1 (T 0 )
LeF
= 1.5 p
.
k11 (T 0 ) k5 (T 0 )
LeO2
(3.6)
The parameters in Eq.(3.6) are the Lewis numbers and rate parameters of important reactions in the
detailed mechanism. The factor C also contains many rate parameters and Lewis numbers. Note that
there is no explicit dependence on the activation temperature in these equations. However, an eective
activation temperature can be derived from:
Ta =
d
4T ad
.
ln s2
L
d(1/T ad )
T ad T 0
(3.7)
37
Although this model qualitatively predicts the structure and propagation of a stoichiometric methaneair ame, it was not able to reproduce measurements and numerical data quantitatively. For this
reason, a number of extensions and improvements have been introduced over the years. In [108] for
instance, Seshadri and Peters introduced so-called fuel consumption and radical consumption layers
inside the inner layer to improve accuracy. Extensions to lean and rich methane-air ames have been
introduced as well. Similar approaches have been proposed to describe the structure and propagation
of some other hydrocarbon ames. Extensions to methane-hydrogen-air mixtures have been introduced by de Goey [39]. Seshadri reviewed the use of asymptotic methods to model ame structure
and propagation in dierent mixtures in [109].
3.2.3
Impact of heat transfer
It is dicult to control free planar ames in the laboratory. The control of such ames, which is for
instance important for measuring the adiabatic burning velocity, has been subject of study for many
years. Perturbations in the ow (e.g. planar stagnation ames [64]), the ame geometry (e.g. spherically expanding ames [41]) and the heat balance (e.g. burner stabilized ames and heat-ux burner
stabilized ames [70]) have been used to control their motion to determine the adiabatic burning velocity. In all these situations, except for the heat-ux method, extrapolations to the unperturbed planar
adiabatic state have to be performed to determine the laminar burning velocity, because the unperturbed ames are unstable. In this subsection, we will consider the control of planar ames by heat
transfer in such a way that they are stabilized on top of or inside porous inert material (PIM). Dierent
materials are used for stabilizing planar ames in practical sustems, such as porous ceramics, sintered
and bre metals, but also perforated ceramics and metals. Flame temperature is not adiabatic anymore
if heat losses are non-zero: T b = T ad q/c p , where T b is the non-adiabatic ame temperature and q
the total heat loss.
It is usually assumed that the porous structure has pores which are much smaller than the ame
scales to enable the use of volume averaged transport equations [5], but it can be questioned whether
this is justied in all of the practical studies performed so far. The one-dimensional volume-averaged
transport equations for ames in PIM contain two coupled equations for the energy of the gas and the
solid including two temperatures, one for the gas T g and one for the PIM T s :
T g
T g
T g
+ uc p,g
g
= S T s T g Q,
t
x
x
x
T s
T s
(1 )s cs
(1 )s
= S T s T g Qrad .
t
x
x
g c p,g
(3.8a)
(3.8b)
A number of additional (material) parameters appears in these equations, such as the solid conductivity
s , density s and specic heat cs , the porosity and the heat transfer coecient per unit volume S .
The equation for the solid also contains the term Qrad because radiative heat transfer is generally very
important. The transport equations for the species are also changed by the introduction of the porosity
and tortuosity tensor [5] to take into account the changes in ow and diusivity due to the changing
ow paths inside the PIM.
The combustion and stabilization of ames interacting with PIM has been studied extensively
using the above set of adapted one-dimensional transport equations. A central result of these studies is
the so-called stabilization diagram, rst introduced by Sathe et al. [111]. This diagram shows how the
burning velocity of the ame structure behaves as a function of the position inside the PIM. In general,
there are three stabilization points in the PIM where the gas ow velocity can become equal to the
local burning velocity. Two points near the upstream boundary of the material and outside the burner
38
near the downstream boundary of the PIM are mostly stable points, while the point inside the PIM
near the outow boundary is unstable. The two stable points indicate two operating modes, already
mentioned by Viskanta [125]: ames stabilized inside and on top of PIM. Transitions between the
stable states are possible for some materials and such situations are studied for instance by Lammers
et al. [61]. These phenomena are referred to as ash-back or blow-o. We will study both operating
modes in the following.
Flames on top of PIM have been studied extensively in the past. Laboratory burners, like McKenna
products burners [87] and heat ux burners [70] are academic burners which use the heat loss of
planar premixed ames to stabilize them. These burners are often used to study the structure of such
ames, for instance using laser-diagnostic techniques like Laser-Induced Fluorescence (LIF). Surface
stabilized ames are also widely used in practical combustion systems nowadays, especially in smallscale house-hold burners and boilers because of their wide operating range in terms of equivalence
ratios and specic powers available to stabilize the ames, their high radiative power and their low
NO emissions. Heat is transferred from the gas to the burner material in the preheating zone of the
ame, where chemical eects are still unimportant. This heat is nally transferred to the environment
by cooling the burner material externally or by radiation (e.g. for ceramic burners), which explains
the lower ame temperatures and low NO emissions.
It has been shown by de Goey [34] that surface stabilized ames behave exactly the same as
adiabatic planar ames propagating in a mixture at a lower initial temperature T u = T u q/c p as
long as the ame temperatures T b = T ad q/c p (q being the heat transferred to the burner), of both
ames are the same and as long as the heat transfer of the ame to the burner only takes place in the
preheating zone of the ame structure, where = 0. For the mass burning rate m of a non-adiabatic
ame with ame temperature T b < T ad we nd
m=
1
c p (T b T u )
Tb
2Q
Tu
dT =
1
c p (T ad T u )
Tb
2Q
dT .
(3.9)
Tu
As a result, it can be shown that the mass burning rate m and the ame temperature T b are eectively
related to each other by
m
1
1
= exp 112T a
.
(3.10)
m0
T b T ad
This relation has already been found experimentally by Kaskan in 1967 [54] and has been veried
by many others afterwards. The heat loss prole of the ame has no inuence on the mass burning
rate as long as it is concentrated in the preheating zone of the ame, as shown by de Goey [34].
This observation has been used by de Goey and his group to develop the so-called heat-ux method
for stabilizing adiabatic ames and for measuring the adiabatic mass burning rate of gas mixtures in a
very accurate way (see for instance [70, 13, 45, 55]). The heat-transfer prole in this situation consists
of a gain part in a cold part of the ame and a loss part in a somewhat hotter region near the outow
of the burner material. The integrated heat transfer S T s T g is 0, inducing an adiabatic situation
and the possibility to stabilize adiabatic ames.
Flames propagating inside PIM have been studied extensively as well (see [123, 47] for recent
reviews). The ame structure changes dramatically compared to the free ame due to the fact that
the heat transfer prole is extended over the complete ame structure and now interacts with the
reaction layer. Generally, it can be stated that heat is gained by the gas from the PIM in the cold part
while heat loss dominates the ame in the burnt gas zone. This leads to so-called super-adiabatic or
excess-enthalpy ames, where ame temperatures are locally higher than the adiabatic value in the
reaction layer even if the complete system is adiabatic [42]. Burning velocities are rising considerably
39
compared to m0 due to two eects: the higher ame temperature T b > T ad and the higher eective
conductivity e > to preheat the mixture in the preheating zone (see Eq.(3.9) where and T b
become higher). In practice, the two PIM boundaries radiate heat to the environment, thereby leading
to heat loss and enabling ame stabilization at lower gas velocities. If the heat transfer is not well
established or controlled, blow-o and ash-back (ignition of the mixture in the upstream part) might
occur. To prevent this from happening, additional layers of materials have been added to the upstream
part of the PIM in practical systems (see for instance [44]) as ame arrestor and/or ow distributor.
As mentioned, the stabilization point near the burner outow in the stabilization diagram is not
stable, forcing the ame to move upstream or downstream to one of the other two stabilization points.
This does not mean that the other stabilization points are always stable. Both operating modes have
shown pulsating solutions. Blackshear et al. [10] and El-Hamdi et al. [29] observed a variety of
pulsating instabilities of ames stabilized on top of a perforated burner, in the form of axial, radial,
drum-head and spiral modes. For the analysis of the axial pulsation a nonlinear model is required if
we are interested in, for example, the ame oscillation amplitude [78, 79], while a linear model is
sucient to study the onset of pulsating instabilities. In analyzing the stability of ames use is often
made of the constant density (or thermal diusive) model [114, 18], so that the mass and momentum
equations are decoupled from the species and energy balance equations. Because of the decoupling,
the pulsating ame problem becomes analytically tractable. However, this is at the expense of the less
realistic constant density assumption. For one-dimensional ames the constant density approximation
is not needed, because the Von Mises transformation allows the density and velocity perturbations
to be eliminated. Using the Von Mises transformation, Margolis [72] and McIntosh and Clarke [82]
derived a dispersion relation using large activation energy asymptotics, which determines under what
conditions a ame becomes unstable. All these models predict that thermo-diusive eects might
lead to pulsations for Le>1. The reason for this is that the enthalpy within the preheating zone due
to Le>1 processes is higher than the value of the ame front. A ame moving in the direction of the
preheating zone due to some distortion, will then be accelerated due to this surplus of enthalpy, leading
to instability. However, these pulsations are seldomly observed in adiabatic ames because the Lewis
number has to be unrealistically large for this to happen [114, 48]. On the other hand, in free ames
that are susceptible to a volumetric heat loss (e.g. by radiation) or in burner stabilized ames, the heat
loss promotes oscillatory instability [18, 48]. In burner stabilized ames oscillations may even persist
if the Lewis number is 1 [82, 72, 50]. Since the Lewis numbers of the major species in methane/air
combustion do not deviate much from 1, pulsations are likely to occur in burner stabilized ames.
All these phenomena are mainly governed by interaction of the heat transfer to the burner with
the ame propagation and are especially manifest at low velocities and at large heat losses, near the
extinction limits. To grasp some of the physics of this process, the following simple model shows
the basics of these phenomena. When a ame oscillates on top of a burner, the heat ux oscillates
also, because of changes in the temperature gradient at the burner. This results in an enthalpy wave
propagating from the burner to the reaction layer and the ame dynamics is inuenced due to the corresponding ame temperature changes and the time lag involved. Consider a reaction sheet stabilized
at x = xf above a perfect burner at x = 0 having a constant temperature T = T u . If the ame is
steady, the mass ow rate u uu equals the mass burning rate of the ame sheet m, where the bar above
a variable denotes its mean value. In case of a temporal variation in the ow rate uu (t) = uu + uu (t), the
ame starts moving and m becomes a function of the space coordinate x and time t. In our model, the
ame is treated as a rigidly oscillating structure only moving without internal changes, i.e., m = m(t),
implicitly assuming that internal ame stretching K = m/x, can be neglected [106]. The question now is: how does the corresponding variation in the mass burning rate m(t) relate to the mass ow
rate changes u uu (t)? If we assume that a change in ame temperature T b leads to an instantaneous
40
change in m according to Eq.(3.10), we nd
u u u
H
m Ze u uu T b Ze
=
=
(f , t),
t
2 T b T u t
2 c p (T b T u ) t
(3.11)
where dH = Q dY + c p dT is the uctuation in the total enthalpy of the system and depends not
only on the time t, but also on the (von-Mises) space coordinate dened by d = dx/u . In the
derivation of (3.11) we assumed that Yb = 0, i.e., fuel is depleted in the ame front at f . Fluctuations
in H are transported with the gas velocity uu , so that heat losses to the burner at = 0 are detected
by the ame front with a time lag f / u . In general, the shape of the enthalpy waves changes due to
u
damping eects. If damping is neglected, and assuming Le = 1, we may write:
H
H
Y
(f , t) =
(0, t f / u ) = Q (0, t f / u ),
u
u
t
t
t
(3.12)
where we used that T = T u is constant at the burner outow. Using the kinematic equation for the
ame motion in terms of Y:
u
Y
Y
(, t) = m(t) u uu (t)
(, t),
t
(3.13)
we nally nd a rst-order dierential-delay equation for m(t):

m(t) Z
Z
+ m(t ) = u uu (t ),
t
f
f
t = t d ,
(3.14)
where we introduced the delay time d = f / u , the ame time f = T / u and the feed-back
u
u
coecient Z given by
Ze T ad T b
Z=
.
(3.15)
2 Tb Tu
Equation (3.14) describes the motion of the ame due to velocity uctuations uu (t ) and will be used
by dr. Kornilov in his part to describe the acoustic transfer function governing the interaction of
acoustic waves with premixed planar burner-stabilized ames. However, also if no acoustic waves
were present (uu = uu ), this equation predicts pulsating solutions, especially if Z is high.
For a harmonic forcing term uu (t) = uu + uu exp(i), the mass ow rate m behaves harmonically,
while m = u uu , and the complete solution of Eq. (3.14) is given by
m(t) = m + meit met/d eit/d ,

(3.16)
where we used the initial condition m(0) = m. Furthermore, and have to satify the relations
= Ze cos(),
= Ze sin(),
(3.17)
where Z = f Z/T .
The homogeneous part of the solution (3.16) indicates when solutions are stable or unstable. Stable solutions are found when < 0, while the solution becomes unstable if > 0, even if there is
no external forcing ( u = 0). Unstable solutions are found when > 0, for which /2 < < and
u
Z > /2. In Fig.2 an example is shown for a burner-stabilized methane/air ame. Flame temperature
T b , stand-o distance given by

f
= ln(T ad T u )/(T ad T b )
(3.18)
T
41
Figure 3.2: Feed-back coecient Z, stand-o distance f and ame temperature T b as function of uu for a
methane/air ame. Parameter values are: = 0.8, T ad = 2016 K, T u = 300 K, T a = 34.566 K and sL = 26 cm/s.
and Z are plotted as function of uu . Note that Z > /2 for velocities lower than approximately 15 cm/s,
indicating instability for velocities u < 15cm/s [72]. In numerical (and experimental) observations,
instability is found for much lower velocities. This indicates that this theory is insuciently accurate
to predict instability. If damping of enthalpy waves is considered, a lower value for the critical velocity is found, in better accordance with experiments and other theoretical models [105]. Qualitative
agreement exists between models, numerical results and experiments. However, quantitative agreement is not accomplished yet due to the sensitivity of the models to small changes, especially if one
relies on simplied kinetics, as shown by Gielen [33]. Flame pulsations are also observed for ames
inside PIMs [61] although they have not been studied in a systematic way.
3.3
Flame stretch
This section deals with the inuence of multi-dimensional eects on the propagation and dynamics of
premixed laminar ames. On the smaller scales of the ame front, the multi-dimensional ame front
can be curved and the approaching ow not homogeneous, for instance leading to local stretching or
compression of the ame front. By introducing a local ame adapted coordinate system, the system
can still be described by a quasi one-dimensional system of equations with additional terms, related
to ame stretching and curvature, reecting transport phenomena along the ame front. This results
in changes in the local burning velocity sL , depending on the local thermo-diusive nature of the
ame front. On the larger scales, ames might display complicated dynamics, invoked by instabilities
from a hydrodynamic nature, damped or enhanced by phenomena with a thermo-diusive character.
This might lead for instance to cellular structures and complicated dynamics of ames in external
force elds, like acoustic waves. This subject is of importance for practical applications using laminar
premixed ames, like Bunsen-type burners and combustors, but also for scientic reasons, because
this knowledge is also the starting point of theories for the dynamics of turbulent ames.
These are the topics of study of this section. In the rst subsection, ame stretch and curvature
phenomena are reviewed and the quasi one-dimensional set of so-called amelet equations is presented. The analysis of this set of equations is analyzed in the subsequent subsection by reviewing
the eects of these distortions on the burning velocity. This behavior is subsequently used in the next
section to study ame stability and to display their eect on ame dynamics.
42
3.3.1

Generalized ame stretch
It is well known that ame stretch, caused for instance by local velocity gradients near the ame front,
may have an important inuence on premixed laminar ames. This was rst recognized by Karlovitz
et al. [52] and Markstein [74]. Flames, experiencing ame stretch have been studied by a lot of
researchers since, such as Buckmaster, Matalon, Matkowsky, Chung and Law [16, 17, 75, 76, 22].
A rigorous mathematical analysis of stretched ames, based on matched asymptotic expansions, was
given by Matalon and Matkowsky [75]. An integral analysis of ame stretch was presented by Chung
and Law [22]. Flame sheet models with one-step chemistry and a single Lewis number are frequently
used to study stretched ames. Stimulated by the ideas of Buckmaster [16], who rst realized that
variations in the preheating zone thickness are of importance and using ideas of the integral analysis
of Chung and Law [22], de Goey et al. [35, 36] applied the ame stretch formalism to ames with
nite ame front thickness.
We base this review on the more fundamental formulation of de Goey et al. [35] and relations with
earlier work will be explained. The governing equations are formulated in a ame adapted coordinate
system, from which a set of quasi-one-dimensional amelet equations arises with extra source terms,
expressed in terms of the ame stretch rate. On the basis of the ame adapted coordinate system,
the full system of equations can be split in three parts as follows [35]: (a) the mass and momentum
equations describing uid ow, (b) the kinematic condition, formulated as a generalized G-equation
[131], describing the ame motion, and (c) the amelet system for the inner ame structure and
burning velocity.
The ame adapted coordinate system is dened in the following way. The ame zone is dened
in terms of some progress variable Y, strictly increasing between the unburned (Yu ) and burned
(Yb ) ame boundaries (see Fig. 3.3). For Y we may take the temperature or any combination of
species mass fractions, provided Y 0 everywhere in the ame. Next, we introduce the curvilinear
coordinate system = 1 , 2 , 3 , referred to as ame coordinates, by the relations
Y(x1 , x2 , x3 , t) = Y 1 ,
x x
=0
1
( = 2, 3)
for Yu Y Yb ,
(3.19)
i.e., coordinate surfaces 1 = Const coincide with ame surfaces Y(x, t) = Const and the 1 coordinate lines are orthogonal to these. The ame surfaces move with velocity vf = v + sL n, where
n is the unit normal (on the ame surfaces) directed towards the unburned gas mixture, resulting in a
time-dependent coordinate system = (x, t). The motion of the ame surfaces satises the kinematic
condition
Y
+ v Y = sL | Y|,
(3.20)
t
which is reminiscent of the G-equation [131]. However, this equation holds in the entire ame zone
and not just on an innitely thin ame sheet.
Due to ow straining and ame curvature, the local ame structure and burning velocity will
change. These eects are included in the stretch rate K [63, 76], dened for a ame surface as
the fractional rate of change of area of an innitesimal surface element dS = d2 d3 , i.e., K =
1/ d/dt. Traditionally, K is dened on the ame front, regarded as an interface separating unburnt
and burnt gases. However, ow and ame distortions might also compress or expand the ame front
in 1 -direction, i.e., it might change the thickness of the ame (see also sections 2.3 and 2.4). As
an extension for ames of nite thickness, the mass-based stretch rate K [35, 36, 121] is introduced,
which is dened as the fractional rate of change of mass contained in an innitesimal volume element
43
Figure 3.3: Visualization of the temperature eld in a premixed ame (left top) with a zoom including the ame
contours (constant Y-values) and amelet paths (perpendicular to ame contours) in a Cartesian coordinate
system (right top) and temperature eld in ame adapted system (bottom).
dV =
reads
g d1 d2 d3 moving with velocity vf . Formulated in ame coordinates, the expression for K
K=
1 dM
1 d
1 dh1 1 d
1 d
=
= K +
+
.
g +
M dt
g dt
dt
h1 dt
dt
(3.21)
Thus, compared to K , K includes additional terms, e.g., due to ame thickness variations (with scale
x
factor h1 =| 1 |= g/ in 1 -direction). It is important to note that K is a scalar eld dened in the
entire ame and not just on an innitely thin ame surface like K .
Combining the denition of K with the mass continuity equation and using the relation vf = v+sL n
we get
(mn) = K,
(3.22)
where we introduced m = sL as the local mass burning rate of the ame iso-plane Y. This equation
states that K is responsible for changes in m along the amelet through the ame. In ame coordinates,
we obtain the following set of conservation equations for mass, species mass fractions and energy
44
(temperature):
m = K,
s
Y
mY
DY
= (Y Kc p Y),
s
s
s
1
Yi
mYi
= (i KYi ), i = 1, ..., Ns ,
s
Lei s c p s
(3.23b)
mT
= (T KT ),
s
s c p s
(3.23d)
(3.23a)
(3.23c)
where ds = h1 d1 is the arclength along the 1 -coordinate lines. In the derivation of system (3.23)
it is assumed that diusive uxes tangent to the ame surfaces are negligible compared to the stretch
terms proportional to K, which is justied in the amelet regime [35]. We can derive similar equations
for the (specic) enthalpy h and the element mass fractions Z j . The solution of this so-called amelet
system (3.23) includes the ame structure in terms of Yi (s), T (s) and the mass burning rate m(s) = sL
of the amelet.
3.3.2
Eects of stretch and curvature on ame propagation
In the present subsection, we analyse the solution of the amelet system, where it is assumed that the
stretch rate K and the curvature are given functions of s. This functional dependence, in fact, is
determined by the local conditions of the amelet in a given ow eld. Note that K(s) is close to a
constant for weakly strained and curved ames in a ow eld with large scale distortions. For strong
stretch and curvature, K(s) and (s) can be wildly varying functions of s. We will derive general
expressions for the change in mass burning rate and also consider some special cases. The focus is,
again, on quantitative physical descriptions of the phenomena.
Analogous to the derivation of (3.9), multiplying the amelet equation (3.23b) with Y/s and
integrating across the ame, both for K = 0 and K 0, we obtain the relation
1+
KaY
mb (b ) = m0 (b ),
b
LeY
(3.24)
for the mass burning rate mb at the burned side of the ame and its stretchless counterpart m0 , where
b
KaY is the Karlovitz integral for Y, given by
KaY =
LeY
(m)b
sb
K Y ds,
su
Y(s) Yu
Y(s) :=
.
Yb Yu
(3.25)
The mass burning rates are assumed to be functions of the state variables b = (hb , Z j,b ), describing
the state of the burnt equilibrium mixture [37]. Note, however, that b as argument of m0 corresponds
b
to a stretched ame and that m0 = m0 0 . For weakly stretched ames we can expand m0 b in a
b
b
b
b
rst order Taylor series around the stretchless state 0 , to obtain
b
mb m0
b
m0
b
KaY
=
+ hb
ln m0 +
b
LeY
hb
Ne
Z j,b
j=1
ln m0 + hot,
b
Z j
(3.26)
45
where hb = hb h0 and Z j,b = Z j,b Z 0 . In order to determine the increments hb and Z j,b ,
b
j,b
we have to integrate the amelet equations for h and Z j across the ame. Assuming that all diusive
uxes vanish at the unburned and burned sides of the ame, we nd [35]
1+
KaY
hb =
LeY
Ns
i=1
Kai KaY 0
h (Yi,b Yi,u ),
Lei
LeY i
(3.27)
where Kai is the Karlovitz integral of the ith species, dened analogous to KaY in (3.25), with Y(s)
i (s). Similar expressions hold for Z j,b . Thus, changes in enthalpy and element mass
replaced by Lei Y
fraction are due to a combination of ame stretch and preferential diusion.
It is important to know at this stage, how accurate the expressions (3.24)-(3.27) are. In [93]
Eq.(3.24) has been used to analyse turbulent ame kernels in the thin reaction zones regime. Even for
very highly stretched and curved amelets, Eq.(3.24) predicts the numerically computed changes in
mass burning rate very accurately for the case that all Lewis numbers Lei = 1. This formulation might
therefore be the starting point of strong stretch theory in the future. Results for the general case Lei 1
are not yet thoroughly compared with detailed numerical results. As a rst result, equation (3.26) in
combination with relation (3.27) for hb and similar expressions for Z j,b have been validated with
numerically computed mass burning rates of weakly strained planar ames with detailed kinetics
and constant Lewis numbers [119]. To evaluate the Karlovitz integrals in the weak stretch limit we
may replace , Y and m in (3.25) by their stretchless equivalents 0 , Y0 and m0 , determined from
the amelet equations (3.23) with K = 0. This comparison has shown that these equations are able to
predict all important changes in element mass fractions, enthalpy and mass burning rate in numerically
computed stretched ames. It seems that the Lewis numbers of all species take part in the preferential
diusion eects. These last equations therefore seem complicated and hard to apply to the general
case of ames described by multiple species chemistry and transport.
We therefore rst consider the simplied case of a stagnation ame, governed by a one-step irreversible reaction F P with a single Lewis number Le and an innitely thin reaction layer [121]. In
this case we nd that Z j,b = 0, leading to
mb m0
b
m0
b
= Mb Kab + O Ka2 ,
b
(3.28)
where the Karlovitz number is given by Kab = b KT /m0 , K = a being the (constant) stretch rate, and
b
Mb is the Markstein number, given by
Mb = ( + 1)
1
1
log( + 1) + 112Ze(Le 1)
Le2 log(1 + ) d ,
(3.29)
0
with = T b T u /T u the thermal expansion coecient. The rst and second term in (3.29) describe
the direct stretch eect and the combined eect of preferential diusion in combination with ame
stretch [37].
An alternative approach to derive (3.28) and (3.29) is presented by Clavin [26]. He considers
the ame as a surface separating the (incompressible) burned and unburned gas mixtures and the
combustion equations are integrated across the reaction layer to give boundary and jump conditions
for the governing equations in the (un)burned mixtures, which are subsequently solved using matched
asymptotic expansions (Ze ). He obtained an equivalent result, except that in the integral in Eq.
(3.29) Le has to be replaced by 1. However, in his derivation it was assumed that Le 1 = O 1/Ze ,
46
so that this result is in lowest order in 1/Ze equal to ours. A similar expression was derived by
Matalon and Matkowski [75]. Rogg and Peters performed a weak stretch analysis for a stoichiometric
methane-air ame described by a systematically reduced 4-step kinetic model and found a similar
expression as Eq.(3.29). The sensitivity of the mass burning rate to the temperature, i.e., the coecient
Ze/2 = (T ad T u ) ln mb /T ad in Eq.(3.29), is replaced by 2(T ad T u )/(T ad T 0 ), similar to Eq.(3.7).
An interesting alternative is based on the intrinsic disturbed ame equations (IDFE) by Class et
al. [23, 24], which is a quasi-one-dimensional system of conservation laws, analogous to the amelet
equations (3.23), although formulated in terms of dierent ame coordinates. The ame itself is
represented by the surface 1 = 0 separating the burned and unburned gas mixtures. They apply IA
to the IDFE, both with and without reaction terms, to derive jump conditions that replace (the eect
of) the reaction terms. Subsequently, solving the IDFE coupled with the jump conditions by matched
asymptotic expansion gives a relation for the laminar burning speed sL , which has to be used in (3.20)
to determine the propagation of the ame.
The question remains how well the expressions (3.24)-(3.27) describe stretch related phenomena
in ames governed by realistic chemistry and multi-species transport. In case of constant Lewis numbers unequal to 1 for all species, Groot et al. [40] have shown that the terms describing the eect
of changes in enthalpy and elements due to dierential diusion, can be rewritten as a combination of terms related to changes in ame temperature and major species products H2 O, CO2 and O2 .
Furthermore, recent studies indicate that the sensitivity of the burning velocity to changes in ame
temperature is dominant. These temperature variations are induced by a combination of changes in
enthalpy and element mass fractions (mainly equivalence ratio). For fuel lean ames, Groot et al.
have shown that all Lewis number except the fuel Lewis number cancel. This then remarkably reduces to the single unit Lewis number expression (3.29) for Mb , where the Lewis number equals the
fuel Lewis number LeFuel . For stoichiometric and rich mixtures, similar expressions are not valid.
One-step expressions using the oxidizer Lewis number or a combination of fuel and oxidizer Lewis
numbers cannot reproduce numerical results for stoichiometric or rich ames as proposed in one-step
chemistry models [49]. Note that Groot et al.[40] have also shown that Markstein numbers of pure
curvature and strain parts in ames might be dierent, but the combined inuence is described by the
above models.
In order to get more insight in the dierent contributions of non-unit Lewis number eects on the
propagation velocity of stretched ames, as an example we present a weak stretch analysis of the mass
burning rate of lean ( = 0.6) and rich ( = 1.4) propane-air ames, computed numerically with a detailed reaction mechanism [110]. In the simulations, one-dimensional weakly-stretched steady ames
(K(s) = a) are computed with constant Lewis numbers for all species. The numerical mass burning
rate mb is shown as function of the Karlovitz number Kab in gure 3.4 with black markers. To understand the impact of dierent assumptions in the Markstein number expressions, several assumptions
in the derivation of (3.29) are relaxed. To start with, we consider the expressions (3.26)-(3.27) valid
for weakly stretched ames, where the burned side of the ame at s = sb is chosen at the location
where the fuel consumption rate has decreased to 5% of its maximum value in the ame. For the
progress variable we choose Y = YC3 H8 . This result is shown in the gure with a black solid curve.
Note that this expression quite accurately reproduces numerical results. On the other hand, black
dashed lines represent the unit Lewis number result of Eq.(3.26) without preferential diusion, i.e.,
hb = Z j,b = 0. The numerical result is not at all reproduced.
The enthalpy and element mass fraction changes can be written in terms of changes in temperature
and mass fractions of the major species H2 O, CO2 and O2 as mentioned. Neglecting the species terms
47
as indicated above and taking only variations in T b into account gives:

Ka0
mb
Ze T b
Y
=1
+
.
0
0
LeY
2 T ad T u
mb
(3.30)
Taking T b from the numerical solution of (3.23) gives the green curves in the gure. This approximation is accurate for both the lean and rich case, indeed. Next, assuming one-step chemistry, the
temperature change can be expressed as the dierence between the Karlovitz integrals of temperature
and the decient (lean) species:
0
Ka0
mb
Ze Kalean
KaT ,
=1
+
Le
0
LeY
2
mb
lean
(3.31)
where lean refers to the Karlovitz integral of the lean species (either fuel or oxidizer). This relation
is plotted with a red curve. It works well for the lean case but not for the rich (as expected). Finally,
assuming analytical (exponential) proles for 0 , T 0 and Yi0 nally gives Eq.(3.29). This expression
is shown in blue and is a very poor representation of the true result. Note that a much more accurate
evaluation can be obtained substituting the one-dimensional numerical solution of (3.23) for , T and
Yi instead of exponential solutions.
1.04
0.95
1.02
0.9
0.85
0.98
y
0.8
0.96
0.75
0.94
0.7
0.92
0.65
0.02
0.04
x
0.06
0.08
0.9
0.01
0.02
0.03
0.04
Figure 3.4: Mass burning rate of stretched propane-air ames. Left: Lean mixture ( = 0.6). Right: Rich mixture ( = 1.4). Black symbols: numerical simulations. Black dashed line: mb /m0 = 1 Ka0 /LeY . Black curve:
b
Y
Equation (3.26). Green curve: Equation (3.30). Red curve: Equation (3.31). Blue curve: Equation (3.29).
Markstein numbers have also been determined experimentally for many mixtures. For instance,
during experiments on laminar burning velocity using the constant volume method, corrections for
ame stretch are needed, which enables the evaluation of Markstein numbers at the same time [41].
Reliability is rather limited though when the corresponding position in the ame front where the
Markstein number has to be evaluated is not accurately known, because its value is very sensitive
48
to this position [122]. More reliable experimental data and a more thorough confrontation with numerical results and analytical expressions are needed. It is recommended to update chemical kinetic
mechanisms and transport models from simultaneously obtained accurate burning velocity and Markstein number data in future studies. Flame stretch in curved ames are not yet thoroughly analyzed
although Eq.(3.24) and Eq.(3.26) are valid for curved ames as well. Extension of the analysis to
curved ames is therefore recommended as well.
3.4
Flame stability
The previous section focussed on the small-scale physical and chemical processes inside the ame
front. In the present section we will consider what happens with the ame on a larger scale. It is well
known that, depending on the Markstein number, ames may show dierent kinds of intrinsic instability. Well-known examples are the formation of cells and ridges in quasi-stabilized and expanding
ames of rich propane/air mixtures. Flame front instabilities have been studied in many papers and
reviews are given in, e.g., [115, 26, 65, 77].
The starting point of the analysis is the well-known theoretical studies of Darrieus [27] and Landau [62], who independently predicted that planar ames are unconditionally unstable due to thermal
expansion of the gas, also known as the hydrodynamic instability. The Landau-Darrieus (LD) theory assumes an innitely thin ame front with a constant burning velocity sL and thermo-diusive
processes inside the front are discarded. If the planar ame is perturbed with alternating convex and
concave segments towards the unburnt mixture, then streamlines are bend towards the ame normal
due to thermal expansion. Therefore, the streamlines must diverge in front of a convex region, resulting in a decreasing local velocity. As a result the burning velocity will be larger than the gas velocity
and the perturbation will grow. Mathematically, the instability is found by expanding the perturbation
in normal modes of the form exp(t + ik), with the growth rate, the transverse coordinate and k
the wave number. Substitution in the conservation equations leads to a dispersion relation = (k).
Without gravity, it reads
(k) = ksL LD ,
LD =
1+
1 + 1/ 1 ,
(3.32)
with = u /b the density ratio. Note that the growth rate is positive for all wave numbers since
> 1. If gravitation is included, this leads to an additional stabilizing part in LD for long wavelengths, depending on the orientation. At smaller wavelengths the hydrodynamic instability will always prevail.
In practice, however, stable planar ames do exist. The reason for this discrepancy is related to
the thermo-diusive processes inside the ame front, characterized by the inuence of ame stretch
on the burning velocity as described in the previous section. When thermal-diusive eects inside the
ame front are taken into account, Eq.(3.32) is extended by additional terms :
(k) = ksL LD k2 sL T B1 + B2 112Ze(Le 1) ,
(3.33)
with Bi > 0, (i = 1, 2) constants [77]. The term with B1 is related to the direct stretch eect and the
B2 -term is related to the preferential diusion eect as discussed in the previous subsection. First, let
us consider equi-diusive ames (Le = 1), for which case the burning velocity always decreases with
increasing stretch rate. Since a convex region of the perturbed planar ame is positively stretched by
ow straining and ame curvature, its burning velocity decreases, which has a stabilizing eect as
seen in Eq.(3.33). For non-equidiusive ames there is an additional eect due to the combination
49
of stretch and the imbalance of diusion of mass and heat. For an eective Lewis number Le > 1,
positive stretch induces a lower ame temperature and an even lower burning velocity; see previous
section. Thus, at a convex region, where K > 0, the gas velocity decreases due to the hydrodynamic
instability, but the burning velocity decreases as well due to direct stretch and preferential diusion
eects. These stretch related eects damp the hydrodynamic instability if the term with T is positive,
or in other words, if the eective Lewis number exceeds a critical value Lec . On the other hand, for
Le < Lec the Markstein number is negative and the short wavelength perturbations are also unstable
and the hydrodynamic instability is enhanced by the thermal-diusive eects. Since this so-called
thermal-diusive instability is caused by a change of the diusive structure of the ame, the cell size
of the instability is expected to be one order larger than the ame thickness. An additional T -term
can be introduced in Eq.(3.33) to describe viscous eects inside the ame front [77]. It should be
noted nally, that heat losses increase the instable combustion regime, like in the case of planar ame
pulsations.
From the previous one can conclude that the eective Lewis number, or equivalently, eective
Markstein number as related by Eq. (3.29), is a key parameter of the thermal-diusive instability. For
a one-step reaction between fuel and oxidizer, this eective Lewis number is found to be a weighted
average of the individual Lewis numbers LeF and LeO2 [49]. It changes with equivalence ratio from
LeF for fuel-lean mixtures to LeO2 for fuel-rich mixtures, indicating that the diusivity of the decient
reactant determines the characteristics of thermal-diusive eects. This seems to agree with eective
Lewis numbers for lean hydrogen and propane mixtures extracted from simulations of symmetrical
counterow ames [118]. For lean multi-component fuel mixtures, Law et al. [66] have shown that
the eective Lewis number reduces to the weighted average of the Lewis numbers of the dierent fuel
components, the weights determined by their contribution to the total heat release. In the previous
section, it has been shown that this model is not valid to predict Markstein numbers in stoichiometric
and rich ames described by multiple species and multiple Lewis numbers. For the propane-air ames
considered in the previous subsection, Markstein numbers are positive and large for lean mixtures but
negative and large (in absolute value) for rich mixtures. The lean ame is therefore predicted to be
stable, while the rich ame is instable. Matalon [77] recently reviewed current models available for
predicting premixed ame stability. Dispersion relations like Eq. (3.33) are only available for some
special cases like a perturbed planar ame and a spherically propagating ame. For more complex
ame shapes, less is known. Furthermore, these expressions explicitly rely on small distortions, while
it is known that nonlinear eects are very important to describe ame dynamics. For more general
cases, numerical models should be used with reliable models for the inuence of small-scale stretch
eects on the mass burning rate as derived in the previous section. This especially holds for ames
in turbulent ows. A very popular model is the G-equation model with a realistic description for
the inuence of the stretch and curvature on ame dynamics [95]. Note that such models are only
valid if all ow scales are much larger than the ame thickness. In case of small-scale perturbations,
inhomogeneous stretch and curvature elds perturb the ame structure and the model breaks down,
including the weak stretch theory. These small-scale perturbations can be taken into account, but
then one has to rely on models where the ame front thickness T is resolved, like so-called laminar
amelet models. An example is the use of amelet-generated manifolds [98], for which it has been
shown that the extended stretch theory of the previous section (Eq.(3.24)-(3.27)) holds.
50
References
[1] S. A. Abrukov, V. V. Kurzhunov and V. N. Mezdrikov, Combustion, Explosion, and Shock Waves
2 (2) (1966) 43-44.
[2] V. V. Afanasev Combustion, Explosion, and Shock Waves 35 (3) (1999) 252-260.
[3] V. V. Afanasev, N. N. Kidin, Moscow, Fizmatlit (in Russian), 2008.
[4] H.M. Altay, S. Park, T. Wu, T. Wee, A.M. Annaswamy, Z.A. Ghoneim Proc. Combust. Inst, 32
(2009) 1359-1366.
[5] J. Baer and Y. Bachmat, Kluwe Academia Publishers, 1990.
[6] F. Baillot, D. Durox, and R. PrudHomme, Combustion and Flame,88(2) (1992) 149-168.
[7] F. Baillot, A. Bourehla, and D. Durox, Combustion Science and Technology, 112 (1996) 327350.
[8] A.L. Birbaud, D. Durox, and S. Candel, Combustion and Flame, 146 (3) (2006) 541-552.
[9] P.L.J. Blackshear, Proc. Combust. Inst. 4 (1953) 553-566.
[10] J.I. Blackshear, J.W. Mapp and M. Gordon Comb. Sci. and Techn., 35 (1984) 311-315.
[11] D.W. Blair, F.C.T. Shen, Combust. Flame 13(4) (1969) 440-442.
[12] J.D.B.J. van den Boom, A. Konnov, A.M.H.H. Verhasselt, V.N. Kornilov, L.P.H. de Goey and H.
Nijmeijer Proc.Combust. Inst. 32 (2009) 1237-1244.
[13] K.J. Bosschaart and L.P.H. de Goey Combust. Flame, 136(3) (2004) 261-269.
[14] J.P. Botha, D.B. Spalding, Proc. R. Soc. Lond. A 225 (1954) 71-96.
[15] R.J. Bowser, F.J. Weinberg, Combust. Flame 18(2) (1972) 296-300.
[16] J.D. Buckmaster, Acta Astronautica 6 (1979) 741-769.
[17] J.D. Buckmaster, Q. J. Mech. Appl. Math. 35 (1982) 249-263.
[18] J.D. Buckmaster, SIAM J. Appl. Math. 43 (1983) 1335.
[19] W.B. Bush and F.E. Fendell, Comb. Sci. and Techn. 1 (1970) 421-428.
[20] S. Candel, Proc. Combust. Inst., 29 (2002) 1-28.
[21] J. M. Cha and S.H. Sohrab, Combust. Flame 106:467-477, 1996.
[22] S.H. Chung and C.K. Law, Combust. Flame 72 (1988) 325-336.
[23] A.G. Class, B.J. Matkowsky and A.Y. Klimenko J. Fluid Mech. 491 (2003) 11-49.
[24] A.G. Class, B.J. Matkowsky and A.Y. Klimenko J. Fluid Mech. 491 (2003) 51-63.
[25] P. Clavin and F.A. Williams, J. Fluid Mech. 116 (1982) 251-282.
51
[26] P. Clavin, Prog. Energy Combust Sci., 11 (1985) 1-59.

[27] G. Darrieus, Propagation dun front de amme, Presented at La Technique Moderne (Paris) and
in 1945 at Congr` s de M canique Appliqu e (1938).
e
e
e
[28] S. Ducruix, D. Durox, and S. Candel, Proc. of Combustion Institute,28 (2000) 765 - 773.
[29] M. El-Hamdi, M. Gordon, J.W. Mapp and J.I. Blackshear Comb. Sci. and Techn., 55 (1987)
33-40.
[30] A.B. Fialkov, Prog. Energy Combust. Sci. 23(5-6) (1997) 399-528.
[31] M. Fleil, M., A.M. Annaswamy, Z.A. Ghoneim, and A.F. Ghoniem, Combustion and Flame
106(4) (1996) 487-510.
[32] J.S. Fox, I. Mirchandani, Combust. Flame 22(2) (1974) 267-268.
[33] A.J.M. Gielen PhD Thesis TU Eindhoven, the Netherlands, ISBN 90-386-2784-X, 2003
[34] L.P.H. de Goey, A. van Maaren and R.M. Quax, Combust. Sci. and Tech., 92(201), 1-3, 1993
[35] L.P.H. de Goey, R.M.M. Mallens and J.H.M. ten Thije Boonkkamp, Combust. Flame, 110 (1997)
54-66.
[36] L.P.H. de Goey, and J.H.M. ten Thije Boonkkamp, Combust.Sci. and Techn., 122 (1997) 399.
[37] L.P.H. de Goey and J.H.M. ten Thije Boonkkamp, Combust. Flame 119 (1999) 253-271.
[38] L.P.H. de Goey and J.H.M. ten Thije Boonkkamp, Combust. Sci. Techn. 122 (1997) 399-405.
[39] L.P.H. de Goey, R.T.E. Hermanns, R.J.M. Bastiaans, Proc. Combust. Inst., 31(1) (2007) 10311038.
[40] G.R.A. Groot, J.A. van Oijen, L.P.H. de Goey, K. Seshadri, N. Peters, Combust. Theory Modelling, 6(4) (2002) 675-695.
[41] X.J. Gu, M.Z. Haqa, M. Lawes and R. Wooley Combust. Flame 121 (1-2) (2000) 41-58.
[42] D.R. Hardesty and F.J. Weinberg, Combust. Sci. and Techn. 8 (1974) 201-214.
[43] A. van Harten, A.K. Kapila, and B.J. Matkowsky, SIAM J. Appl. Math. 44 (1984) 982.
[44] C. Hayashia, I. Malico and J. C. F. Pereira Computers and Structures 82 (2004) 1543-1550.
[45] R.T.E. Hermanns, A. Konnov, R.J.M. Bastiaans and L.P.H. de Goey, Laminar Burning Velocities
of Diluted Hydrogen-Oxygen-Nitrogen Mixtures, Energy and Fuels, 21(4) (2007) 1977-1981.
[46] B. Higgins, J. Natural Phil. Chem. Arts 1 (1802) 129.
[47] J. R. Howell, M. J. Hall and J. L. Ellzey Prog. Energy Comb. Science 22 (2) (1996) 212-145.
[48] G. Joulin and P. Clavin, Combust. Flame, 35 (1979) 139-153.
[49] G. Joulin, T. Mitani, Combust. Flame 40 (1981) 235-246.
52
[50] G. Joulin, Combust. Flame, 46 (1982) 271-282.

[51] R. Kaess, T. Poinsot, and W. Polifke Proc. of the European Combustion Meeting, Vienna, 2009.
[52] B. Karlovitz, D.W. Denniston, D.H. Knappascharefer, and F.E. Wells, Fourth Symposium on
Combustion, Williams and Wilkins, Baltimore, 613, 1953.
[53] N. Karimi, N., M.J. Brear, S.H. Jin and J.P. Monty, Combustion and Flame 156 (2009) 22012212.
[54] W.E. Kaskan 6th Int. Symposium on Combustion, the Combustion Institute (1967) 134-143.
[55] A. Konnov, I.V. and Dyakov, Combust. Sci. and Tech., 179 (2007) 747-765.
[56] V.N. Kornilov, PhD thesis TUe Eindhoven, 2006.
[57] V.N. Kornilov, K.R.A.M. Schreel, and L.P.H. de Goey, Proc. Combust. Inst, 31 (2007) 12391246.
[58] V.N. Kornilov, K.R.A.M. Schreel, and L.P.H. de Goey in Proc. ICSV14, 2007.
[59] V.N. Kornilov, M. Manohar, and L.P.H. de Goey, Proc. Combustion Institute32 (2009) 13831390.
[60] V.N. Kornilov, R. Rook, J.H.M. ten Thije Boonkamp, L.P.H. de Goey,
Flame156 (2009) 1957-1970.
Combustion and
[61] F.A. Lammers and L.P.H. de Goey, Combust. Flame, the Combustion Institute, 133 (1-2) (2003)
47-61.
[62] L. D. Landau, Acta Physicochim. USSR 19 (1944) 7785.
[63] C.K. Law, Proc. Comb. Inst. 22 (1988) 1381-1402.
[64] C.K. Law, Springer Verlag, Berlin 1993.
[65] C. K. Law, C. J. Sung, Prog. Energy Combust. Sci. 26 (2000) 459-505.
[66] C. K. Law, G. Jomaas, J. K. Bechtold, Proc. Combust. Inst. 30 (2005) 159167.
[67] Chung K. Law, Combustion Physics, Cambridge University Press, Cambridge, 2006.
[68] J. Lawton, F.J. Weinberg, Claredon Press, Oxford, 1969.
[69] Lieuwen, T.C., Journal of Propulsion and Power19(5) (2003) 765-781.
[70] A. van Maaren and L.P.H. de Goey Combust. Sci. and Tech. 102 (1994) 309-314.
[71] E. Mallard and H.L. Le Chatelier, Ann. Mines 4 (1883) 274-399.
[72] S.B. Margolis, Combust. Sci. Techn., 22 (1980) 143-169.
[73] Markstein, G.H., Proc. Combust. Inst. 3 (1984) 162-167.
[74] G.H. Markstein, Nonsteady Flame Propagation, Pergamon Press, Oxford, 1964.
53
[75] M. Matalon, and B.J. Matkowsky, J. Fluid Mech. 124 (1982) 239-259.
[76] M. Matalon, M., Combust. Sci. and Techn. 31 (1983) 169-181.
[77] M. Matalon, Proc. Comb. Inst. 32 (2009) 57-82.
[78] B.J. Matkowsky and D.O. Olagunju, SIAM J. Appl. Math., 39(2) (1980) 290-300.
[79] B.J. Matkowsky and D.O. Olagunju, SIAM J. Appl. Math., 40(3) (1981) 551-562.
[80] Y. Matsui, Combust. Flame, 43 (1981) 199-209.
[81] A.C. McIntosh and C.F. Clark, AIAA Progress in Astronautics and Aeronautics, 88 (1982) 3.
[82] A.C. McIntosh and J.F. Clark, Combust. Sci. Tech. 38 (1984) 161-196.
[83] A.C. McIntosh, Proc. Inst. Acoustics, 6 (1984) 277-298.
[84] A.C. McIntosh, Combust. Sci. Tech. 49 (1986) 143-167.
[85] A.C. McIntosh, Combust. Sci. Tech., 54 (1987) 217-236.
[86] H.J. Merk, Proc. Combust. Inst.,6 (1956) 500-512.
[87] F. Migliorini, S. de Iuliis, F. Cignoli and G. Zizak Combust. Flame 153 (2008) 384-393.
[88] M. Munjal, Wiley, New York 1987.
[89] N. Noiray, D. Durox, T. Schuller and S. Candel Combustion and Flame 145(3) (2006) 435-446.
[90] Noiray, D. Durox, T. Schuller and S. Candel, Proc. Combust. Inst31 (2007) 1283-1290.
[91] N. Noiray, D. Durox, T. Schuller, and S. Candel, J. Fluid Mech., 615 (2008) 139-167.
[92] N. Noiray, D. Durox, T. Schuller and S. Candel, Proceedings of the Combustion Institute, 32
(2009) 3163-3170.
[93] J.A. van Oijen, R.J.M. Bastiaans, G.R.A.Groot, L.P.H. de Goey, Flow, Turbulence and Combustion, 75 (2005) 67-84.
[94] N. Peters and F.A. Williams, Combust. and Flame, 68 (1987) 185 207.
[95] N. Peters, Turbulent Combustion, Cambridge, 2000.
[96] B. Rogg and N. Peters, Combust. and Flame, 79 (1990) 402-420.
[97] J. A. van Oijen, L. P. H. de Goey, Combust. Theory Modelling 6 (2002) 463-478.
[98] J.A. van Oijen, G.R.A. Groot, R.J.M. Bastiaans, L.P.H. de Goey, Proc. Combust. Inst., 30 2005,
657-664.
[99] T. Poinsot and D. Veynante,
Philadelphia, 2005.
Theoretical and Numerical Combustion, 2nd ed., Edwards,
[100] S.H. Preetham and T.C. Lieuwen, in 43rd AIAA Aerospace Sciences Meeting and Exhibition,
2004.
54
[101] A.A. Putnam, Combustion Driven Oscillations in Industry, New York: Elsevier, 1971.
[102] R.L. Raun, M.W. Beckstead, J.C. Finlinson, and K.P. Brooks, Prog. Energy Combust. Sci. 19
(1993) 313-364.
[103] J.L. Rayleigh, Nature, 18 (1878) 319.
[104] P.L. Rijke, Phil. Mag. 17 (1859) 319.
[105] R. Rook, PhD Thesis, Eindhoven University of Technology,
http://alexandria.tue.nl/extra2/200112945.pdf, 2001.
[106] R. Rook, L.P.H. de Goey, L.M.T. Somers, K.R.A.M. Schreel, R. Parchen, Combust. Theory
Modelling, 6(2): 223-242, 2002).
[107] T. Schuller, D. Durox, and S. Candel, Combustion and Flame 134(1-2) (2003) 21-34.
[108] K. Seshadri and N. Peters, Combust. and Flame, 81 (1990) 96-118.
[109] K. Seshadri Proc. Comb. Inst., 26 (1996) 831-846.
[110] San Diego Mechanism, March 2005, http://maeweb.ucsd.edu/ ombustion/cermech/.
c
[111] S.B. Sathe, R.E. Peck and T.W. Tong, National heat transfer conference HTD, 106 (1989)
461-468.
[112] K.R.A.M. Schreel, R. Rook, L.P.H. de Goey, Proc. Combust. Inst., 29 (2002) 115.
[113] G.I. Sivashinsky, Acta Astronautica 3 (1976) 889-918.
[114] G.I. Sivashinsky, Combust. Sci. Techn., 15 (1977) 137-146.
[115] G. I. Sivashinsky, Ann. Rev. Fluid Mech. 15 (1983) 179-199.
[116] C. Sondhauss, Pogg. Ann. Phys. Chem. 79 (1850) 1.
[117] T. Sugimoto, and Y. Matsui, Proc. Combust. Inst. 19 (1982) 245-250.
[118] C. J. Sun, C. J. Sung, L. He, C. K. Law, Combust. Flame 118 (1999) 108-128.
[119] J.A.M. de Swart, G.R.A. Groot, J.A. van Oijen, J.H.M. ten Thije Boonkkamp, L.P.H. de Goey,
Combust. Flame, 145(1-2) (2006) 245-258.
[120] J.H.M. ten Thije, L.P.H. de Goey, J.A. van Oijen, A.G. Class and Y. Bronner, Combust. Sci and
Techn., 180 (2008) 1449-1477.
[121] J.H.M. ten Thije and L.P.H. de Goey, J. Eng. Math., 62 (2008) 67-84.
[122] J.H. Tien and M. Matalon, Combust. Flame, 84 (1990) 238-248.
[123] D. Trimis, O. Pickenacker, K. Wawrzinek in Cellular Ceramics, M. Scheer and P. Colombo
(Ed.) Wiley-VCH Verlag, 2005
[124] Trun, K. and T. Poinsot, Combustion and Flame142(4) (2005) 388-400.
55
[125] R. Viskanta, 8th Int. Symp. on Transport Phenomena in Combustion, 1995.

[126] E.N. Volkov, A. Sepman, V.N. Kornilov, A.A. Konnov, Y. Shoshyn and L.P.H. de Goey, Proc.
European Combustion Meeting, Vienna, Austria, 2009.
[127] E.N. Volkov, V.N. Kornilov and L.P.H. de Goey, Proc. ICSV16, Krakow, Poland, 2009.
[128] E.P. Weinaas and J. McChesney, Combustion and Flame, 15 (1) (1970) 85-87.
[129] F.J. Weinberg, Uitgever London Academic Press, 1986.
[130] F.A. Williams, Combustion Theory, 2nd ed., 1985.
[131] F.A. Williams, In The Mathematics of Combustion, (J. Buckmaster, Ed.), Society for Industrial
and Applied Mathematics, Philidelphia, (1985) 97-131.
[132] F.A. Williams, Ann. Rev. Fluid Mech., 3 (1971) 171-188.
[133] D.L. Wismana, S.D. Marcuma and B.N. Gangulyb, Combustion and Flame 151 (2007) 639648.
[134] Ya.B. Zeldovich, G.I. Barenblat, V.B. Librovich and G.M. Makhviladze, The Mathematical
Theory of Combustion and Explosions, 1985.
56
Chapter 4
Numerical Modelling of Laminar Flames:

a Concise Introduction
J.H.M. ten Thije Boonkkamp
4.1
A mathematical model for laminar ames
In this section we present a mathematical model for laminar ames. We give a summary of available
numerical methods and briey address the diculties associated with the numerical simulation of
laminar ames.
The governing equations of a laminar ame are the conservation equations of mass, momentum
and energy of the gas mixture and the balance equations of mass for the N s species involved. These
equations read [22, 29]:
+
t
(Yi ) +
t
(v) +
t
(h) +
t
(vYi ) +
(v) = 0,
(vv) = p +
T + g,
(4.1)
(vh) = q,
(Yi Vi ) = wi ,
(i = 1, 2, . . . , N s 1).
The dependent variables in these equations are the density , ow velocity v, pressure p, specic
enthalpy h, temperature T and species mass fractions Yi . Other variables in (4.1) are the viscous stress
tensor T , gravitational acceleration g, heat ux vector q, diusion velocities Vi and reaction rates wi .
The rst two equations in (4.1) are referred to as the ow equations and the latter two as combustion
equations. Appropriate initial and boundary conditions for (4.1) have to be specied. Moreover, the
conservation equations (4.1) have to be completed with the caloric equation of state
Ns
Yi hi ,
h=
hi := h0 +
i
i=1
c p,i () d,
(4.2)
T0
which denes h as a function of the temperature T and the species mass fractions Yi , and the thermal
equation of state
RT
p0 =
,
M
1
:=
M
57
Ns
i=1
Yi
.
Mi
(4.3)
58
Chapter 4 J.H.M. ten Thije Boonkkamp
In equation (4.2), hi , h0 and c p,i are the specic enthalpy, specic enthalpy of formation at reference
i
temperature T 0 and specic heat at constant pressure of species i, respectively. In (4.3), R is the
universal gas constant, M the average molar mass of the gas mixture and Mi the molar mass of species
i. The pressure p is set to a constant value p0 in (4.3), which is a valid approximation for low Mach
number ow [7].
Models for T , q, Vi and wi are required. The gas mixture in a laminar ame behaves like a
Newtonian uid and therefore the viscous stress tensor T = (k,l ) is given by
k,l =
vk vl
+
123( v)k,l ,
xl xk
(4.4)
where is the viscosity coecient of the mixture. An often employed model for the diusion velocities Vi is the so-called generalized law of Fick [19], i.e.
Yi Vi = Di,m Yi ,
(i = 1, 2, . . . , N s 1),
(4.5)
where Di,m is the mixture-averaged diusion coecient, describing the diusivity of species i in the
mixture. Relation (4.5) holds for trace species. Moreover, the diusion velocity of the abundant
species, numbered N s , follows from the constraint
Ns
Yi Vi = 0.
(4.6)
i=1
For the heat ux we use the common expression [29]

Ns
q = T +
hi Yi Vi ,
(4.7)
i=1
where is the thermal conductivity. This expression for q only takes into account enthalpy transport
through conduction and mass diusion. Models for the reaction rates wi are reported in literature; see
e.g. [28].
Using (4.3), we can express the density as a function of the combustion variables T and Yi .
Inserting this function into the continuity equation and combining the resulting equation with the
combustion equations, we nd the constraint
v = S,
(4.8)
where S describes expansion of the gas mixture due to conduction, diusion and heat production, see
[26]. Equation (4.8) is called the expansion equation and is used in Section 4.4 to derive a method for
the pressure computation.
The numerical simulation of the above mathematical model generally involves the following steps:
1. space discretization
2. time integration
3. solution of algebraic systems
4. post processing.
Numerical Modelling of Laminar Flames
59
In the space discretization step, we cover the domain with a grid and approximate all spatial derivatives
on this grid somehow. There are various classes of methods available, the most important of which
are the nite element methods (FEM), nite dierence methods (FDM) and nite volume methods
(FVM). All these methods are frequently used in combustion theory and each method has its merits
and drawbacks. We will not consider FEM; for an introductory text see e.g. [3]. For time dependent
problems, we get after space discretization a set of ordinary dierential equations, referred to as
discrete conservation laws. The next step is then to integrate these equations. The main issue here is
stability of the time integration technique. Next we have to solve the resulting nonlinear system. The
most frequently used method for this is Newton iteration with all its variants. Furthermore, Newton
iteration requires the solution of large linear systems. There is a host of methods available for the
numerical solution of linear systems. Which method should be used depends on the properties of the
linear system involved. The last step in the numerical simulation procedure is post processing of all
data to make plots or movies of a ame. We will not discuss this topic.
Special attention should be given to the computation of the pressure. Unlike all other variables, the
pressure p does not satisfy a convection-diusion-reaction equation, it only occurs as a source term in
the momentum equations. Moreover, it can not be computed from an equation of state, as is the case
for compressible ows. This makes that the pressure only occurs implicitly in the discrete system. A
class of methods to compute the pressure are the so-called pressure correction methods (PCM). In a
PCM, the pressure computation is decoupled from the velocity computation in a predictor-corrector
fashion. For incompressible ows, these methods are well established; see e.g. [9]. However, for
combustion problems these methods are more dicult to apply and a lot of problems still have to be
solved.
The above mathematical model for laminar ames has the following characteristics which makes
it dicult to solve numerically. First, the conservation laws (4.1) are nonlinear partial dierential
equations of convection-diusion-reaction type. Complex ows are possible and the chemical source
terms are often extremely nonlinear. Consequently, Newton iteration has to be modied in order to
converge at all. Second, the number of species and chemical reactions is often large, which make
numerical simulation prohibitively expensive. To circumvent this problem, reduction techniques are
often applied; see e.g. [?, 20]. Third, a laminar ame has quite dierent time and length scales. As an
example, gure 4.1 shows the mass fractions of some species and the temperature of a at methane/air
ame. In this gure, we can distinguish the burnt and unburnt gases, where the solution is virtually
constant, and a very narrow region, the so-called ame front, where the actual combustion occurs and
where all variables rapidly change. It is intuitively clear that inside the ame front much more grid
points are required than in the (un)burnt gases, to get an accurate representation of the solution.
We have organized the remainder of this paper as follows. In Section 4.2 we discuss FVM and
FDM for space discretization of the conservation equations (4.1). Time integration of the resulting
discrete conservation equations is the topic of Section 4.3. In Section 4.4, we briey address a PCM for
the computation of the pressure. Finally, in Section 4.5, we discuss solution methods for (non)linear
algebraic systems.
4.2
Space discretization
We discuss in this section FVM and FDM for space discretisation of the conservation laws (4.1) and
the expansion equation (4.8). Special attention should be given to the momentum equations.
All conservation laws in (4.1) can be written in the generic form
() +
t
f = s,
f := v ,
(4.9)
60

0.2
2500
0.18
0.16
2000
0.14
0.12
1500
0.1
0.08
1000
0.06
0.04
500
0.02
Figure 4.1: Mass fractions of major species (left) and temperature [K] (right) of a one-dimensional methane/air
ame. On the horizontal axis is the distance shown in centimeters.
for some variable , where f is the ux and s a source term. Finite volume methods are based on the
integral form of the conservation law in (4.9), i.e.
d
dt
d +
V
f n d =
s d,
(4.10)
which holds for an arbitrary subvolume V with boundary V and unit outer normal n. In fact, equation (4.10) is the most fundamental conservation law, which reduces to the partial dierential equation
in (4.9), provided the variable is smooth enough.
In nite volume methods, the whole domain is covered with a grid consisting of a nite number
of control volumes and the integral form (4.10) is applied to each of these control volumes. For the
sake of simplicity, we restrict ourselves to a two-dimensional, rectangular and uniform grid. A patch
of such a grid is shown in gure 4.2. We see that the grid nodes are located at the centres of the
control volumes. Applying the integral form (4.10) to the control volume VC corresponding to the
central point C and approximating all integrals involved by the midpoint rule [6], we nd the discrete
conservation law
d
1
1
(Fe Fw ) +
(Gn Gs ) = sC .
()C +
(4.11)
dt
x
y
In (4.11), ()C denotes the approximation of at grid point C and Fe denotes the approximation
of the x-component f = u of the ux at the centre of the eastern cell face e, etc. The nite
x
volume method has to be completed with the computation of the numerical uxes Fe , Gn , etc.
A numerical ux should satisfy the following three requirements. First, it should be second order
accurate for both diusion dominated as well as convection dominated problems, since the corresponding discrete conservation law (4.11) is also second order accurate in x and y. Second, the re-
61
NW
NE
Figure 4.2: Patch of a nite volume grid with position names.
SW
SE
sulting scheme may not cause unphysical oscillations in the numerical solution and, third, the scheme
may only have a three-point coupling in each spatial direction.
Numerical uxes are usually computed with a FDM. Consider the computation of the numerical
ux Fe . There are several possibilities for its computation. A rst obvious possibility is the central
dierence approximation, given by
Fe = 112(u)e (C + E )
e
(E C ) .
x
(4.12)
This approximation is second order accurate, however, it produces unphysical oscillations when convection is dominant; see e.g. [8]. A remedy to this shortcoming is the upwind numerical ux
Fe = e (max(ue , 0)C + min(ue , 0)E )
e
(E C ) .
x
(4.13)
This ux is based on a one-sided dierence approximation of the convective term and consequently, it
is only rst order accurate. There exist higher order upwind approximations of the ux, see e.g. [16,
17], however the resulting schemes have more than a three-point coupling in each spatial dimension.
Another possibility is the exponential ux approximation due to Patankar [21]. In this case, the
numerical ux is computed from a local, one-dimensional conservation law without source term. This
way we nd
e
e
(B(Pe )E B(Pe )C ) ,
Fe = F
, Pe ; C , E :=
(4.14)
x
x
where B(z) := z/(ez 1) is the Bernoulli function and Pe := (u)e x/e the local Peclet number.
In fact, the resulting exponential scheme is a combination of the central dierence and the upwind
schemes. For Pe = 0 it is identical with the central dierence scheme and for |Pe | it reduces to
the rst order upwind scheme. A nice generalization of the exponential ux approximation is by van t
Hof et al. [12]. In this case the numerical ux is computed from a local one-dimensional conservation
law including the source term. The result is
Fe = F
P,e ,e
, Pe ; C , E + 112 W(Pe ) xsC ,
Pe x
(4.15)
62
ne
C
s
Figure 4.3: Patch of a staggered grid with control

volumes for ue and vn .
se
where W(z) := (ez 1 z)/(z(ez 1)) is a weight function and v,e := W(Pe )vC + W(Pe )vE (v = P, )
a weighted average of the variable v. A further advancement, taking into account the cross ux term
1
y (G n G s ), is also possible; for more details see [12]. The resulting scheme is second order accurate,
even for convection dominated ows, and does not generate unphysical oscillations.
Special attention should be given to the momentum equations. Consider e.g. the momentum
equation for the velocity component u in x-direction, for which s = p , i.e. the pressure gradient is a
x
source term driving the ow. Omitting for simplicity the viscous terms, the discrete conservation law
(4.11) changes to
d
1
u2 u2
(u)C +
e
dt
x
1
1
(uv)n (uv)s =
(pe pw ) .
y
x
(4.16)
Assume that the left hand side of this equation is known. Since the pressure pe at the eastern cell
boundary is unknown, we have to interpolate it from adjacent grid values, i.e. pe = 112(pC + pE );
and likewise pw = 112(pW + pC ). Substituting these relations into the right-hand side of (4.16), we
obtain
1
1
(pe pw ) =
(pE pW ) .
(4.17)
x
2x
We see that there is no direct connection between neighbouring values of p and the consequence of this
is a decoupling of the pressure eld. To be more precise, if pi, j (t) denotes the numerical approximation
of the pressure in gridpoint xC = (xi , y j ) at time t, then it is clear that also pi, j (t) := pi, j (t) + P(t)(1)i ,
for an arbitrary function P(t), satises relation (4.17). Ignoring any boundary conditions, also pi, j (t)
is a solution of the discrete x-momentum equation (4.16) and we say that the grid function (1)i is
a spurious mode of the numerical pressure. Also taking into account the y-momentum equation, we
conclude that the numerical pressure has the follwing three spurious modes: (1)i , (1) j and (1)i+ j .
These modes are of course not physical and should be avoided.
A way to avoid spurious pressure modes is to use a staggered grid. In a staggered grid, the control
volumes for u and v are displaced as shown in gure 4.3. Consider again the x-momentum equation.
If the viscosity coecient is constant, it can be written in the standard form (4.9), with the ux f
and source term s given by
f = vu u,
q
s = , q := p 113 v.
x
(4.18)

Taking into account that
q
x
=
d
dt
63
(qe x ), we nd the corresponding integral conservation law

u d +
V
f n d =
qn x d,
(4.19)
with n x = n e x the x-component of the unit outer normal on V. Applying this integral form to the
control volume Ve corresponding to ue and approximating all integrals involved by the midpoint rule,
we nd
d
1
1
(FE FC ) +
(Gne Gse ) =
(u)e +
dt
x
y
1
(pE pC ) +
( v)E ( v)C , (4.20)
x
3x
where FE denotes the approximation of the x-component of the ux f = u2 u at the eastern

x
cell face E, etc.; see gure 4.3. Note that equation (4.20) contains the pressure values pC and pE
at adjacent grid points, thus excluding spurious modes in the numerical pressure. Furthermore, in
the staggered grid, the arrangement of grid points is such that locations for all uxes are located
precisely halfway between grid points for the corresponding variables. Consequently, we can apply
all previously mentioned ux approximations, without using any interpolations.
The discretization of the expansion equation (4.8) is based on the integral form
v n d =
S d,
(4.21)
applied to the control volume VC ; see gure 4.2. Approximating the integrals in (4.21) by the midpoint
rule, we nd
1
1
(ue uw ) +
(vn vs ) = S C ,
(4.22)
x
y
where S C = S (xC ), ue is the numerical approximation of the velocity component u at the eastern cell
face, etc. Note that ue can be interpreted as a ux with = 1 and = 0.
The space discretization procedure has to be completed with the implementation of the boundary
conditions. Usually, satises either a Dirichlet boundary condition
= ,
(4.23)
or a Neumann boundary condition
= .
(4.24)
n
As an example, we consider the control volume VC adjacent to the boundary, as shown in gure 4.4.
Since there is no grid point on the boundary, we introduce a virtual grid point as indicated in the gure.
Using linear interpolation, we can easily show that the Dirichlet boundary condition (4.23) gives
112 (C + V ) = (xe ).
(4.25)
Likewise, the Neumann boundary condition (4.24) is approximated by

1
(V C ) = (xe ).
x
These relations can then be used to elimate V from the discrete conservation laws.
(4.26)
64
Figure 4.4: Control volume adjacent to a boundary.
All codes for the numerical simulation of laminar ames have the possibility to adaptively rene
the grid in the vicinity of the ame front. The general approach is to subequidistribute some weight
function W, which is a measure for the smoothness of the solution, on the grid, i.e.
W d eps,
(4.27)
where C are appropriately chosen grid cells and eps is a user specied constant. Based on such a
subequidistribution, grid cells are rened. There are several rening techniques, resulting in truly
unstructured grids or locally uniform grids. An example of the rst renement strategy is the Local
Rectangular Renement method by Bennett and Smooke [4, 5] and an example of the latter is the
Local Defect Correction method by Anthonissen [1]. For more details, the reader is referred to these
publications
4.3
Time integration
After space discretization of the conservation laws in (4.1), we obtain a set of ordinary dierential
equations, which has to be integrated in time. For a comprehensive survey on integration methods for
ordinary dierential equations, see e.g. [10]. Key issue in the choice of a time integrator is stability of
the method. For this reason, we have to restrict ourselves to implicit time integration methods.
Time scales in a laminar ame range typically from 109 s, for the very fast processes, to 102 s,
for the slowest processes, which makes the system of dierential equations very sti [?]. If we want
to simulate all the temporal details, we have to choose a time step compatible with the fastest time
scales and we might then use an explicit method. This would require an extremely large number of
time steps making numerical simulation not feasible. In practical ame simulation, we typically want
to resolve time scales of approximately 1 ms, resulting in a few thousand time steps at most. In
such a simulation we can resolve all processes (ow, transport etc.) except the fastest reactions.
To demonstrate that we have to use implicit time integration methods, we consider the following
standard equation
2
= 2,
(4.28)
t
x
modeling conduction and diusion in a ame. If we discretize this equation, using the central dierence approximation for the second derivative and the explicit Euler method for time integration, we
obtain the scheme
1 n+1
n =
n 2n + n ,
(4.29)
j
j
j1
t j
x2 j+1
65
with x the grid size, t the time step and n the numerical approximation of (x j , tn ). To investigate
j
stability of this scheme, we substitute the Fourier mode
n = n eikx j ,
j
(4.30)
with the amplication factor and k the wave number. Ignoring initial and boundary conditions, the
Fourier mode (4.30) is a solution of (4.29), if is given by
= 1 4d sin2 112 ,
:= kx,
(4.31)
with d := t/x2 . Stability of scheme (4.29) means that (rounding) errors remain bounded during
time integration, and a necessary condition for this is that || 1 for 0 < . This stability
condition leads to the time step restriction
t
x2
.
2
(4.32)
In practice, time step restriction (4.32) is often also a sucient condition for stability. For a typical
gas ame, we need a minimum grid size x = 106 m to resolve the reaction layer and we have
= 3 m2 /s for hydrogen radicals. Combining these data with the time step restriction in (4.32), we
nd a maximum time step
tmax 1013 s = 1010 .
(4.33)
Consequently, we have to choose our time step t very much smaller than the desired time scale ,
because of stability. On the other hand, if we apply the implicit Euler method for time integration, we
obtain the scheme
1 n+1
j n =
n+1 2n+1 + n+1 ,
(4.34)
j
j
j1
t
x2 j+1
which has the amplication factor
= 1 + 4d sin2 112
(4.35)
Obviously, || < 1 and therefore the implicit Euler scheme is unconditionally stable. This means
that we have no time step restriction and that we can choose our time step according to the desired
resolution in time.
Basically, every implicit time integration method is unconditionally stable and can therefore be
used in numerical ame simulations. A drawback of the implicit Euler method is that it is only
rst order accurate in the time step t. Therefore, the method is suitable to compute steady state
solutions, where time accuracy is not of importance, but for transient problems, such as oscillating
ames, higher accuracy is desirable. A nice alternative in this case could be diagonally implicit
Runge-Kutta methods [10]. Both the implicit Euler method and the Runge-Kutta methods are socalled one-step methods, since they include only two time levels. Alternatively, there exist implicit
multi-step methods, which are highly accurate. However, these schemes have more than two time
levels, which makes them very expensive in terms of memory requirements.
Time integration is often combined with operator splitting or time splitting. The general idea of
operator splitting is to break up the semi-discrete conservation law in smaller parts and subsequently
solve the dierent parts using the most appropriate time integration methods. Thus we can choose the
time integration method which is most suitable for a particular subsystem. As an example, consider
the semi-discrete approximation of the conservation law in (4.9), i.e.,
d
+ Fad () = Fc (),
dt
(4.36)
66
where Fad () is the (nite volume) discretisation of the advection and diusion terms ( f) and where
Fc is the approximation of the chemical source term s. To advance the solution from time level tn to
tn+1 we could subsequently solve the following initial value problems
d
+ Fad ( ) = 0, (tn ) = n ,
dt
d
= Fc ( ), (tn ) = (tn+1 ),
dt
(4.37a)
(4.37b)
i.e., we rst solve the advection-diusion system using for example a mixed explicit-implicit time
integrator, a so-called IMEX method, and subsequently solve the (sti) system (4.37b) using a sti
integrator. There are many variants of this approach; for a detailed account see [15].
4.4
A pressure correction method
In this section we give an outline of a PCM to decouple the pressure computation from the computation
of the ow velocity and the combustion variables. For a more detailed discussion, the reader is referred
to [13].
Applying a nite volume space discretization method to the momentum equations, the combustion
equations and the expansion equation, we nd a system of dierential algebraic equations (DAEs),
which can be symbolically written as
dv
= C(, v)v G()p,
dt
d
= A() + B()v,
dt
(4.38)
Dv = S(),
where the vectors v, p and contain approximations of the ow velocity, pressure and combustion
1
variables, respectively. The terms G()p and Dv represent approximations of p and v, respectively. The pressure vector p is a source term in the discretized momentum equations, and it has to be
determined such that the discrete expansion equation is satised at all time levels.
We can derive an explicit equation for p in the following way. Dierentiating the discrete expansion equation with respect to t and combining the result with the rst dierential equation in (4.38),
we obtain
d
L()p = DC(, v)v S(), L() := DG().
(4.39)
dt
1
Note, that the term L()p represents an approximation of p . The pressure equation (4.39) is
sometimes referred to as the hidden constraint, to contrast it with the expansion equation which is a
distinct constraint.
Based on the discussion in Section 4.3, we choose the implicit Euler time integration method.
This way, we nd the following discrete system
vn+1 t C(n+1 , vn+1 )vn+1 G(n+1 )pn+1 = vn ,

n+1 t A(n+1 ) + B(n+1 )vn+1 = n ,
Dvn+1 = S(n+1 ),
(4.40)
67
where vn+1 is an approximation of v(tn+1 ), etc. System (4.40) is generally dicult to solve for vn+1 ,
pn+1 and n+1 simultaneously, due to the hidden constraint for the pressure. To circumvent the solution
of (4.40), we introduce a two-stage PCM.
In the two-stage PCM , system (4.40) is solved in a predictor-corrector fashion. In the predictor
step, the pressure vector pn+1 is replaced by the pressure vector pn at the previous time level. This
way, we nd the following system
v t C(n+1 , v)v G(n+1 )pn = vn ,
n+1 t A(n+1 ) + B(n+1 )v = n ,
(4.41)
for the predicted velocity vector v and the vector of combustion variables n+1 . Note that the predicted
velocity v does not satisfy the discrete expansion equation. In the corrector step, the velocity vector
vn+1 in the term C(n+1 , vn+1 )vn+1 in (4.40) is replaced by the predictor value v, i.e.
vn+1 t C(n+1 , v)v G(n+1 )pn+1 = vn .
(4.42)
Subtracting the rst equation in (4.41) from (4.42) we obtain the relation
vn+1 v = tG(n+1 )pn ,
(4.43)
for the pressure dierence pn := pn+1 pn . Next, applying the discrete divergence operator D
to (4.43) and imposing the expansion equation, we nd the following system for pn+1 :
1
Dv S(n+1 ) ,
t
= pn + pn .
L(n+1 )pn =
pn+1
(4.44)
After we have computed pn from the rst equation in (4.44), we can compute vn+1 from the simple
relation
vn+1 = v tG(n+1 )pn ,
(4.45)
which follows readily from (4.43). Summarizing, we rst compute v and n+1 from the coupled
system (4.41). Next, we compute the pressure pn+1 from (4.44) and nally we compute the new
velocity vn+1 from (4.45). This PCM can be modied by applying a few corrector steps, however, one
corrector step is usually sucient.
4.5
Solution methods for algebraic systems
After space discretization, and if necessary time integration, we obtain a large nonlinear system of
algebraic equations. This system has to be solved iteratively and involves the solution of linear systems
at each iteration step. In this section, we will outline numerical methods to solve nonlinear and linear
systems occuring in numerical ame simulation.
Consider the nonlinear system (4.41), which we have to solve in the predictor step of the pressure
correction method. This system can be symbolically written as
F(y) = 0,
(4.46)
where the vector of unknowns y contains the predicted velocity v and the combustion variables n+1 ;
see Section 4.4. However, system (4.46) might equally well represent some other subproblem in our
68
mathematical model; think e.g. of the discretization of one of the combustion equations. Virtually all
solution methods for (4.46) are based on Newtons method. It is dened by
y(0)
given,
y(k+1) = y(k) J1 y(k) F y(k) ,
k = 0, 1, 2, . . . ,
(4.47)
where y(k) denotes the kth iterand and J(y) := F (y) is the Jacobi matrix of F(y). The iteration (4.47)
is repeated until ||F y(k) || < tol||F y(0) || for some user specied tolerance tol. For combustion
problems, iteration (4.47) will almost never converge, unless the initial guess y(0) is very close to the
solution of (4.46). Therefore, we have to modify the Newton iteration (4.47), to improve the initial
guess y(0) and/or the convergence behaviour.
For time dependent problems, an obvious choice for the initial guess y(0) is the solution at the
previous time level. When the time step t is suciently small, Newton iteration should converge. A
way to improve the initial guess is by continuation. Assume that the nonlinear system (4.46) depends
on a certain parameter, lets say the equivalence ratio . Suppose, we have the solution for a value
0 and we want to compute a new solution for a value 1 . We can do this by solving a sequence of
problems where is successively increased from 0 to 1 ; for more details see e.g. [2].
To improve the convergence behaviour of iteration (4.47), we can multiply the update vector
(k) := J1 y(k) F y(k) by a damping factor , with 0 < 1. This way we obtain the damped
y
k
k
Newton iteration
y(0)
(k+1)
given,
= y(k) k J1 y(k) F y(k) ,
k = 0, 1, 2, . . . .
(4.48)
A suitable choice for the damping factor is e.g. k = 1 + ||F y(k) || . This means that when (4.48)
is converging, it reduces for k to the standard Newton method.
Also damped Newton iteration is often not convergent for combustion problems. A way to overcome this problem is to combine Newton iteration with time-stepping. Therefore, consider the system
dy
= F(y),
(4.49)
dt
instead of (4.46). In this formulation, F(y) represents the discretization of all terms in the conservation
laws, except the accumulation terms. We now apply the implicit Euler time integration method to
(4.49) and solve the resulting nonlinear system for z := yn+1 by Newton iteration. This way we get
the iteration
z(0) = yn ,
z
(k+1)
=z
(k)
J z
(k)
1
I
t
F z(k)
1 k
z yn
t
k = 0, 1, 2, . . . .
(4.50)
Note that for t iteration (4.50) changes to standard Newton iteration. The time step t should
1
be chosen such that the matrix J z(k) t I is not singular and not even nearly singular. For stationary
problems, we have to apply several time steps until we reach the steady state solution of (4.49). In
this case we speak of the method of false transient.
Another way to speed up convergence of (4.47) is based on the embedding of (4.46) in the following time-dependent problem
dy
= J1 (y)F(y),
d
y(0) = y(0) ,
( > 0)
(4.51)
69
where is an articial time. The dierential equation in (4.51) is called the Davidenko equation and its
steady state solution is the solution of (4.46). The damped Newton iteration (4.48) can be considered
as the explicit Euler method applied to (4.51) with time step = k . Improving convergence of (4.48)
is now equivalent with improving stability of the explicit Euler method for (4.51). A way to achieve
this is to apply the mixed Euler method to (4.51), i.e. take J1 (y) at level = k and F (y) at level
= k+1 . This way we obtain
y(k+1) = y(k) k J1 y(k) F y(k+1) ,
k = 0, 1, 2, . . . .
(4.52)
Note that the implicit Euler method would require the inverse Jacobi matrix J1 (y) at the new level
= k+1 , making the method unfeasible. Equation (4.52) is a nonlinear system for z := y(k+1) and can
be symbolically written as
G(z) := z + k J1 y(k) F(z) yk = 0.
(4.53)
This system in its turn has to be solved by Newton iteration, and this way we obtain the iteration
z(l+1) = z(l) I + k J1 y(k) J z(l)
G zl ,
l = 0, 1, 2, . . . .
(4.54)
To reduce the computational costs of (4.54), we set J z(l) = J y(k) during the iteration and we then
obtain
k
1
z(l+1) =
z(l) J1 y(k) F z(l) +
y(k) , l = 0, 1, 2, . . . .
(4.55)
1 + k
1 + k
The complete iteration consists of the outer loop (4.52) and the inner loop (4.55) and has usually a
much better convergence behaviour than standard Newton iteration.
In combustion problems, the Jacobi matrix J(y) is often very dicult to compute analytically,
because of complicated reaction rates and/or complicated transport models for conduction and diusion. For that reason, the Jacobi matrix is usually computed numerically. The entry Ji, j can be e.g.
computed by one-sided dierences as follows
Ji, j :=
Fi 1
Fi (y + e j ) Fi (y) ,
y j
(4.56)
where e j is the jth unit vector and a small positive number; see e.g. [25]. This Jacobi matrix can be
used in any of the iterative methods above.
A major drawback of Newton iteration and all its variants is that the computation of the Jacobi
matrix, either analytically or numerically, is very expensive. Therefore, the Jacobi matrix is often
kept constant during a few or all iterations. Another way is to build up a cheap approximation B(y)
of J1 (y) during iteration using so-called rank-1 updates. The resulting method is called the Broyden
method; see e.g. [11, 14]. All these methods are referred to as quasi-Newton methods.
To compute the pressure update pn from (4.44) but also at each step of Newton iteration or one
of its variants, we have to solve a linear system, which can be symbolically written as
Ay = b.
(4.57)
The eciency of a numerical model highly depends on reliable and ecient solvers for (4.57). We
rst note that the coecient matrix A is sparse, which means that each row in the matrix has only a
few nonzero entries. The precise form of A depends on the numbering of the unknowns and equations.
We rst consider a one-dimensional at ame. Suppose, we collect all physical variables per
control volume, such as e.g.
ue , T C , Y1,C , . . . , YNs 1,C , pC ,
(4.58)
70
C
W
C
W
C
W
S
S
S
S
S
S
S
N
N
Figure 4.5: Sparsity pattern of the Jacobi matrix for a one-dimensional (left) and two-dimensional (right) ame
simulation. In the right matrix some of the W and E blocks are zero blocks corresponding with control volumes
adjacent to the boundary.
and subsequently number the control volumes. If we number the corresponding equations in the same
way, the matrix A has a sparsity pattern as shown in gure 4.5. In fact, A is a block-tridiagonal
matrix corresponding with the three-point coupling in the spatial direction. Each nonzero block in
this matrix represents the coupling between all variables in a control volume. System (4.57) can be
very eciently solved using block-LU factorization, i.e. the matrix A is is factorized as A = LU,
where L and U are a lower and an upper block triangular matrix, respectively. In fact, L has only one
block-subdiagonal and U only one block-superdiagonal. Using this factorization of A, the solution
method for (4.57) reads
solve by forward substitution Lz = b,
solve by backward substitution
Uy = z.
(4.59)
This method is very ecient because L and U together have the same number of nonzero entries as
the original matrix A. Note that the computation of the LU-factorization only has to be done when a
new Jacobi matrix is computed.
The situation is dierent for a two-dimensional ame. Suppose, we want to compute a ame on a
rectangular domain. If we group all physical variables per control volume, number the control volumes
along grid lines parallel to the x-axis and number the corresponding equations in the same way, we
obtain for A a sparsity pattern as shown in gure 4.5. The sparsity pattern corresponds with the vepoint coupling S-W-C-E-N between the variables as, indicated in gure 4.2. Alternatively, the ninepoint coupling SW-S-SE-W-C-E-NW-N-NE would lead to a matrix with nine block-diagonals. The
coecient matrix A in gure 4.5 is a block-pentadiagonal matrix, however, not all block-diagonals
are adjacent. LU-factorization of A would give matrices L and U which have their bands completely
lled. For this reason LU-factorization is expensive in terms of CPU-time and memory requirements.
Consequently, we have to use iterative methods in combination with preconditioning, to solve (4.57).
71
There are many iterative methods available for linear systems; for an excellent survey see e.g. [23].
Reliable methods, suitable for nonsymmetric linear systems, are GMRES [24] and Bi-CGSTAB [27],
and these are most frequently used in ame simulations.
References
[1] Anthonissen, M.J.H., Local Defect Correction Techniques: Analysis and Application to Combustion, Ph. D. Thesis, Eindhoven Univeristy of Technology, 2001.
[2] Ascher, U.M., Mattheij, R.M.M. and Russel, R.D., Numerical Solution of Boundary Value Problems for Ordinary Dierential Equations, Prentice Hall, Englewood Clis, 1988.
[3] Axelsson, O. and Barker, V.A., Finite Element Solution of Boundary Value Problems: Theory
and Computation, Academic Press, London, 1984.
[4] Bennett, B.A. and Smooke, M.D., Local rectangular renement with application to axisymmetric
laminar ames, Combust. Theory Model., 2: 221258, 1998.
[5] Bennett, B.A. and Smooke, M.D., Local rectangular renement with application to nonreacting
and reacting uid ow problems, J. Comput. Phys., 151: 684727, 1999.
[6] Burden, R.L. and Faires, J.D., Numerical Analysis, Brooks/Cole Publishing Company, Pacic
Grove, 1997.
[7] Buckmaster, J.D., An introduction to combustion theory, in: The Mathematics of Combustion,
SIAM, Philadelphia, 1985, pp. 346.
[8] Feistauer, M., Mathematical Methods in Fluid Dynamics, Longman Scientic & Technical, Harlow, 1993.
[9] Gresho, P.M. and Sani, R.L., Incompressible Flow and the Finite Element Method, John Wiley,
Chichester, 1998.
[10] Hairer, E. and Wanner, G., Solving Ordinary Dierential Equations II: Sti and DierentialAlgebraic Problems, Springer, Berlin, 1996.
[11] Heath, M.T., Scientic Computing, An Introductory Survey, McGraw Hill, Boston, 2002.
[12] Van t Hof, B., Ten Thije Boonkkamp, J.H.M. and Mattheij, R.M.M., Discretization of the
stationary convection-diusion-reaction equation, Numer. Methods Partial Dierential Eq. 14:
607625, 1998.
[13] Van t Hof, B., Ten Thije Boonkkamp, J.H.M. and Mattheij, R.M.M., Pressure correction for
laminar combustion simulation, Combust. Sci. and Tech. 149: 201233, 1999.
[14] Van t Hof, B., Numerical Aspects of Laminar Flame Simulation, Ph.D. Thesis, Eindhoven University of Technology, 1998.
[15] Hundsdorfer, W. and Verwer, J.G., Numerical Solution of Time-Dependent Advection-DiusionReaction Equations, Springer Series in Computational Mathematics 33, Springer, Berlin, 2003.
72
[16] Van Leer, B., Towards the ultimate conservative dierence scheme. V. A second order sequel to
Godunovs method, J. Comp. Phys. 32: 101136, 1979.
[17] Leonard, B.P., A stable and accurate convective modelling procedure based on quadratic upstream interpolation, Comp. Meths. Appl. Mech. Eng. 19: 5998, 1979.
[18] Maas, U. and Pope, S.B., Simplifying chemical kinetics: intrinsic low-dimensional manifolds in
composition space, Combust. Flame 88: 239264, 1992.
[19] Majda, A. and Lamb, K.G., Simplied equations for low Mach number combustion with strong
heat release, in: Dynamical Issues in Combustion Theory, Eds. Fife, P.C., Linan, A. and
Williams, F.A., Springer-Verlag, New-York, 1991, pp. 167211.
[20] Van Oijen, J.A., Flamelet-Generated Manifolds: Development and Application to Premixed
Laminar Flames, Ph.D. Thesis, Eindhoven University of Technology, 2002.
[21] Patankar, S.V., Numerical Heat Transfer and Fluid Flow, McGraw-Hill, New York, 1980.
[22] Poinsot, T. and Veynante, D., Theoretical and Numerical Combustion, Edwards, Philadelphia,
2005.
[23] Saad, Y., Iterative Methods for Sparse Linear Systems, PWS Publishing Company, Boston, 1996.
[24] Saad, Y. and Schultz, M.H., GMRES: a generalized minimum residual algorithm for solving
nonsymmetric linear systems, SIAM J. Sci. Stat. Comput., 7, 856869, 1986.
Ph.D. Thesis, Eindhoven University of Technology, 1994.
[26] Ten Thije Boonkkamp, J.H.M., Rook, R. and Mattheij, R.M.M., A pressure correction local
defect correction scheme for laminar ame simulation, Proceedings of ECCOMAS CFD 06,
Egmond aan Zee, The Netherlands, 5-8 September 2006.
[27] Van der Vorst, H.A., Bi-CGSTAB: a fast and smoothly converging variant of Bi-CG for the
solution of nonsymmetric linear systems, SIAM J. Stat. Comput., 12, 631644, 1992
[28] Warnatz, J., Maas, U. and Dibble, R.W., Combustion, Springer Verlag, Berlin, 1996.
[29] Williams, Forman A., Combustion Theory, Addison-Wesley, Redwood City, 1985.
Chapter 5
Reduced chemical models

L.M.T. Somers
5.1
Introduction
In a previous lecture detailed chemical models are discussed. Such a chemical model has in principle
the virtue to be independent of the ame-type which is studied. Formally, the same chemical model
can be used for any type of reacting ow, e.g. diusion and premixed ames either in pseudo onedimensional congurations (at ame, counter-owing jets) or multi-dimensional ows. However,
carrying out calculations is hampered by an important drawback. Due to the stiness present in the
governing equations, implicit techniques have to be used. This leads to large sparse matrix equations
to be solved, with the leading dimension proportional to N K, where N is the number of species
and K the number of mesh-points. It can readily be seen that for modelling of multi-dimensional
reacting ows the memory requirements are large. Furthermore, the Newton-type iteration which is
normally applied invokes a Jacobian matrix which, mostly, is evaluated numerically. This is a very
costly process and is in fact the most important drawback of the detailed chemical models. Only
due to the fast increase in computing power, the simulation of multi-dimensional reactive ows with
detailed chemical models comes within our grasp. However, for the development of engineering tools
for designing concepts as well as for computations of large scale reacting ow congurations, the
application of detailed models is still (and will remain in the near future) impractical1 . Reduction of
the problem size and removing the stiness in the equations is the best way to proceed.
5.2
Global reaction models
In the early years of computational reactive uid dynamics when even simulations with detailed chemistry on the simplest geometries (1D laminar) where impossible due to lack of adequate computer
capacity, so-called global reaction models were used frequently. These models replace the complex
chain reactions with an overall reaction of the type,
Fuel + Oxygen
Products.
Generally the parameters A, b, Ea in the Arrhenius-like reaction rate relation

f u = A[Fu] [Ox] T + eEa /RT
Mid nineties with a contemporary supercomputer the computing time is of the order of two hours for a co-owing
axi-symmetrical laminar diusion ame (computational domain 2.5 cm 30 cm, 15 species reaction model) [1].
73
74
were tted to reproduce data, like the adiabatic burning velocity (S L ) and temperature T ad and occasionally the non-adiabatic ame temperature2 obtained by experiments, e.g. [2, 3]. Later detailed
chemistry simulations, e.g. [4], on 1D ames were used as well. The main advantage of this simplied
chemical model is the reduced stiness of the system of equations. Explicit techniques can now be
used which do not require much storage [2].
The ultimate goal in the systematic reduction techniques in one way or another is reducing the
stiness and degrees of freedom required for application of detailed chemical models yet keeping the
predictive quality with respect to the required physical phenomena. In the subsequent part of this
lecture we will describe the approach, introduced some years ago by Peters et al. [5], referred to as
convential systematically reduced chemical models. It is used in many applications nowadays, see
e.g. [6, 7, 12, 13]. In the nal section a representant of whats called the mathematical systematic
reduction technique is shortly described.
The strategy of the conventional approach is to use the detailed chemical model as basis and
to apply steady-state and partial equilibrium assumptions to eliminate a number of the species in the
model and to replace the corresponding (partial, ordinary) dierential equations by algebraic relations. This leads to a reduction of the dimension of the matrix equations, which is desirable from a
computational point of view. In general a systematically reduced chemical scheme is not very specic for the type of ame considered, e.g. for the one-dimensional premixed and diusion ames the
same model can be used [6, 7]. However, as for the global mechanism, each reduced scheme is to
a certain extent deduced for the mixture which is to be considered and the properties which have to
be predicted accurately. For example, the same basic detailed chemical3 describes the combustion of
most simple hydrocarbons in diusion as well as pre-mixed ame types. However, each ame type
yield a dierent systematically reduced scheme [7].
In remaining part of this lecture we will deal with is adequately referred to as systematical reduction techniques. A deduction of a reduced chemical model is not straightforward, and can be based on
knowledge and experience of the author or on rigourous mathematics. The former will be addressed
as Conventional and the latter as Mathematical.
5.3
Convential systematic reduction technique: mechanism for lean methane-air

ames
Systematically reduced chemical mechanisms are, contrary to earlier mentioned global mechanisms,
obtained by a careful analysis of a detailed reaction model describing the oxidation of a specic fuel.
Eectively, the objective of such an analysis will be to replace the (partial) dierential equations for
some species by algebraic expressions which, in principle, reduces the computational eort. The
reduction strategy is based on two distinctive tasks. First, there is the actual reduction of the number
of dierential equations which have to be solved, i.e. the identication of the species that to adequate
accuracy can be assumed to be in steady-state. Secondly, there is the so-called truncation of the
expressions for the species which are to be considered in steady-state. Both the reduction of the
problem size as well as the truncation of the resulting expressions will be treated in this lecture.
However, rst some general concepts will be introduced which can be applied to any detailed reaction
mechanism. These principles will be applied to the skeletal mechanism used for lean methane-air
ames as already presented in lecture 2, section 2.4 on page 26.
2
Burner-stabilized ames.
E.g. the GRI mechanism [8] can be used for the combustion of several small hydrocarbon molecules (methane,ethane,
propane).
3

5.3.1
75
General principles in systematic reduction
Two important features in the systematical reduction of the detailed chemical models are the so-called
steady-state approximation and the partial-equilibrium approximation.
Steady-state approximation
The most important assumption is the steady-state approximation. It simply states that for a species
the chemical source term is zero, i.e. for the i-th species
M
i = Mi
i j q j = 0,
(5.1)
j=1
with Mi the molar mass. Equation (5.1) gives an algebraic equation for each species considered to
be in steady-state (involving many reactions in general). Obviously, the nomenclature stems from
the description of homogeneous reacting systems, where this equation implies that the time-derivative
of the concentration is zero. In a ame, the assumption implies that the time scales of the reactions
consuming and producing a steady-state species are much shorter than the time scales associated
with diusion and convection. Generally, if no further simplications are made, the assumption (5.1)
for each of the species taken in steady-state form a set of non-linear coupled algebraic equations. To
reduce the complexity of this set and even more to obtain an explicit4 expression, often the partialequilibrium assumption is applied.
Partial equilibrium approximation

In contrast to the steady-state assumption this is an approximation applied to a specic reaction. If
applied to a certain reaction j symbolically denoted by A1 + A2
B1 + B2 the partial-equilibrium
assumption yields,
qj = 0
(5.2)
and the species involved in such a reaction relate to each other through,
[B1 ][B2 ]
kf
= K = r.
[A1 ][A2 ]
k
(5.3)
It is worth mentioning that in the special case that the reaction rate of a steady-state species involves
only one reaction, it eectively is a partial equilibrium approximation for that specic reaction.
The next subsection discusses which species can be considered in a steady-state. The rst part
will deal with the arguments leading to the decisions as they are applied by most authors. The partialequilibrium assumption will be used in the last subsection 5.3.4 where the derivation of an explicit
expression from the set of algebraic equations is derived.
By explicit is meant; the equation that determines a variable is expressed in known quantities only, e.g. Xl = f (Xi ) with
all Xi known at the point where Xl is evaluated.
4
76
5.3.2
Deduction of a reduced mechanism
The success of reducing a detailed chemical model is strongly related to the ability of applying the
steady-state assumptions to the proper set of species. Classically, this is done by simply comparing the
contributions of the dierent parts (convection-, diusion- and reaction- part) in the balance equation
for a specic species and a detailed investigation of the chemical source term. In order to make a
direct comparison between the magnitude of the dierent components in these balance equations and
the reaction rates of the individual reactions q j in the chemical source terms, the balance equations
are often expressed in terms of the specic molenumber5 of species i,
i = Yi /Mi = Xi / M .
The reason for this is best illustrated if we consider the conservation equation, e.g equation (2.23).
For convenience we dene the operator L as
L{Yi } = v Yi
(Dim Yi ).
(5.4)
Whenever the ame is stationary, the time derivative vanishes and the conservation equation (2.23)
reduces to,
M
L{Yi } = Mi
i j q j
(5.5)
j=1
where the expression for i is written explicitly in terms of the elementary reactions. It is clear that
the equation for i yields,

M
L{i } =
i j q j
(5.6)
j=1
where Mi cancels in the right-hand side. This cancelation is the reason for the introduction of i . It
is now possible to compare the magnitude of diusive and convective uxes expressed in i with the
reaction rates q j directly.
The idea behind the steady-state assumption is best illustrated by comparing the magnitude of the
dierent contributions to the balance equation. These terms are evaluated from results of a complete
reaction model specically for two species e.g. H2 O (gure 5.1), generally a non-steady state species,
and HCO (gure 5.2) an intermediate species that will be considered to be in a steady-state later on.
To compare the importance of each term in the convection-diusion equation (5.6) the dierent uxes
and the net chemical reaction source term are shown in the gure on the left as a function of x. On the
right the contributions to the chemical source term in terms of the elementary reactions of the skeletal
mechanism are given.
From Figs. (5.1) and (5.2) it is clear that for H2 O the contribution of the elementary reactions is
comparable to the magnitude of the net reaction rate as well as the magnitude of the convective and
diusive parts. For HCO, on the other hand, the magnitude of the reaction term is almost negligible
compared to that of the elementary reactions of which it is composed. Since the diusive and convective uxes are comparable in magnitude to the net reaction rate in the ame zone their magnitude
is thus negligible compared to that of the elementary rates. One could say that the concentration of
HCO is determined mainly by the balance of formation and consumption reactions, thus implying the
steady-state assumption, equation (5.1). The time scales associated with the consumption and production of HCO are much smaller than the ones associated with diusive and convective processes.
5
Some authors refer to it as the specic abundance of a species i.
77
Source
q (mole cm-3 s-1)
mole cm-3 s-1
Source
Convection
r15
r3
r12
Diffusion
r4
x (cm)
x (cm)
Figure 5.1: The diusion and convection uxes in terms of and the net reaction rate for H2 O (left). The
dierent parts of the source term (right), solid lines indicate the elementary reactions and the dotted line the net
reaction rate. The listed numbers correspond to the ones in table 2.1 on page 30 of lecture 2.
r15
Convection
Diffusion
x (cm)
q (mole cm-3 s-1)
(mole cm-3 s-1)
Source
r14
Source
r16
r17
x (cm)
Figure 5.2: The diusion and convection uxes in terms of and the net reaction rate for HCO are given (left).
The dierent parts of the source term (right), solid lines indicate the elementary reactions and the dotted line
the net reaction rate. The listed numbers correspond to the ones in table 2.1 on page 30 of lecture 2.
78
Original
Weighted
Original
Weighted
CH3
0.36330
0.43496
H
0.71120
2.70910
CH3 O
0.00078
0.00077
O
0.33250
0.38998
CH2 O
0.26770
0.26322
OH
0.74700
0.85958
HCO
0.00852
0.00845
HO2
0.01613
0.01551
CO
4.75700
4.75742
H2 O2
0.00602
0.00575
H2
1.89400
5.18866
Table 5.1: Original and weighted maximum mole-fractions times 100 for an atmospheric adiabatic methane-air
ame = 0.80.
We will now derive a 4-step mechanism for lean premixed methane-air ames. The number of
steady-state assumptions that has to be employed can be obtained by a simple count of the free parameters in the ame problem. The skeletal mechanism can be considered as a 15 parameter (species)
model, since there are 15 independent variables, i.e. the species. In order to obtain a 4 parameter
model out of a set of 15, one has to pose 11 relations. Since conservation of elements is valid as
long as nuclear reactions do not take place, there are 3 conservation equations, one for each of the
elements C, H and6 O. This means that 8 relations have to be introduced articially (the steady state
assumptions). For which species these assumptions apply best will be discussed in the following
subsection.
Selection of the steady-state species

Which species are most suitable for the steady-state assumption can be obtained either by plotting
gures like gure 5.2 or looking at the maximum mole-fraction of each species in the ame for which
the reduced mechanism is to be used. In our case we consider a lean adiabatic premixed ame and
compute the maximum mole numbers of the intermediate species in table 5.1.
The set of intermediates can now be subdivided in two groups, one with a maximum mole fraction
well below 1% and one well above. It is then common practice to consider the rst group to be in
steady-state since, due to fast reactions involving these species, every reaction creating a particle in
this group is eectively balanced with a reaction consuming it. Therefore, the maximum mole fraction
remains small and the net reaction rate can considered to be zero. In table 5.1 weighted mole fractions
are given as well. These are equal to the original mole fraction weighted by the quantity MN2 /Mi,N2 ,
where
2Mi MN2
(5.7)
Mi,N2 =
Mi + MN2
which is roughly proportional to the binary diusion coecients of each species with respect to Nitrogen (which is abundant). It is known from asymptotic analysis that in the reactive layer, diusive
transport is dominant over convective transport [5]). Therefore, the weighted mole fractions are considered to be more signicant and it is better to use the weighted maximum mole numbers. This is
illustrated by the observation that in most treatments [H] is not taken in steady state whilst [O] and
[OH] are. This would not have been concluded if the original maximum mole numbers would have
been considered.
6
Without N2 chemistry the N conservation can be left out since N2 itself acts as a conserved scalar.

5.3.3
79
Actual reduction of the chemical set
In equation (5.8) the dierential equations are given and each individual reaction is specied explicitly.
This only done so for illustrative purposes and numbering is according to the reaction mechanism
presented in 2.4 on page 30. For all species present in the skeletal mechanism this yields,
L{CH4 } = q10 q11 q12
0 = L{CH3 } = q10 + q11 + q12 + q13 q18
0 = L{CH3 O } = q18 q19 q20
0 = L{CH2 O } = q13 q14 q15 + q19 + q20
0 = L{HCO } = q14 + q15 q16 q17
L{CO } = q9 + q16 + q17
L{CO2 } = q9
L{H2 } = q2 q3 + q7 + q11 + q14 + q16 + q19 + q25
L{H } = q1 + q2 + q3 q5 q6 q7 + q9 + q10 q11 +
(5.8)
q13 q14 q16 + q17 q19 + q20 q24 2q25

L{O2 } = q1 q5 + q7 + q8 q18 + q21
0 = L{O } = q1 q2 + q4 q13 + q18
0 = L{OH } = q1 + q2 q3 2q4 + 2q6 q8 q9 q12 q15 + 2q22 q23 q24
0 = L{HO2 } = q5 q6 q7 q8 2q21 + q23
L{H2 O } = q3 + q4 + q8 + q12 + q15 + q23 + q24
0 = L{H2 O2 } = q21 q22 q23
L{N2 } = 0
For the steady-state species the net reaction rate is taken equal to zero. The set is now reduced by
eliminating precisely as much reaction rates as there are steady-state species. For instance if reaction
21 is to be eliminated, the steady-state assumption for H2 O2 can be used to perform this task. Using
q21 q22 q23 = 0, each occurrence of q21 can consequently be replaced by q22 + q23 , thereby
eliminating q21 from all equations. Which reactions are to be eliminated is arbitrary7 and does not
aect the results of a computation as long as the expressions for the steady-state species are not
truncated. The main idea is to eliminate the fastest reactions that destroy the steady-state species
since the slowest are rate-determining. A kind of main chain by which the complete combustion takes
place is then derived. Here, we choose to follow the approach of Peters and Seshadri or Peters and
Williams [10, 11, 5]. From the H2 -O2 subsystem they eliminate reactions q2 , q3 , q7 and q22 being
the fastest consumption rates for O, OH, HO2 and H2 O2 , respectively. For the intermediates in the
C1 -chain, CH3 , CH3 O, CH2 O and HCO, these fastest reactions are q13 , q20 , q14 and q17 . The last
four complete the minimum set of reactions that can be removed from the main reaction chain. Other
reactions, which are linear combinations of the previously selected reactions, disappear as well from
the main chain. For instance, reaction r8 can be written as a linear combination of r3 and r7 i.e.,
H2 + OH
H + HO2
OH + HO2
7
H2 O + H
H2 + O2
H2 O + O2
r3
r7
+
r8 .
See for instance [5, 9] for typical dierences in choices as applied by dierent authors.
80
Similar arguments apply for reactions r4 (= r3 r2 ) and r15 (= r14 + r3 ), leading to the elimination of
reaction rates q4 and q15 .
The resulting work is to carry out the linear algebra with the set of balance equations. This is a
rather tedious job, which can be automated as is done by G ttgens et al. (REDMECH [14]) or the
o
algorithm presented in appendix B of [15]. Both the package developed by G ttgens et al. and the
o
algorithm developed here, only need the basic detailed mechanism, the chosen steady-state species
and for each of the latter a corresponding reaction to be removed from the main chain. The program
then gives the global mechanism and the corresponding overall reaction rates. Here we will present
the results only, yielding the following conservation equations for the non steady-state species,
L{CH4 } = qI
(5.9)
L{CO } = qII + qI
(5.10)
L{CO2 } = qII
(5.11)
L{H2 } = 4qI + qII + qIII qIV
(5.12)
L{H } = 2qI 2qIII + 2qIV
(5.13)
L{O2 } = qIV
(5.14)
L{H2 O } = qI qII + 2qIV
(5.15)
Here, the global reaction rates of the 4-step model are introduced,
qI = q11 + q10 + q12
qII = q9
qIII = q5 q10 + q16 q18 + q19 q21 + q23 + q24 + q25
(5.16)
qIV = q1 + q6 + q18 + q21 q23 .

The so-called principal rate of each overall reaction is the one appearing rst after the equal-sign. The
magnitudes of the overall reactions are determined by these reactions mainly. This can be determined
by a sensitivity analysis. The global reactions are obtained from the stoichiometry in the set (5.95.15). The rate of reaction II, associated with qII , for instance, appears with 1 in the CO-balance
equation (5.10), +1 in the CO2 -balance equation (5.11), 1 in the H2 -balance equation (5.12), +1 in
the H2 O-balance equation (5.15) and is not present in all the others. When we gather all species with
negative numbers on the left hand side and all those with positive numbers on the right hand side, we
nd global reaction II,
CO + H2 O
CO2 + H2
(rII )
In a similar fashion all other overall reaction may be obtained and are presented below
CH4 + 2H + H2 O
CO + 4H2
(rI )
CO + H2 O
CO2 + H2
(rII )
H+H+M
H2 + M
(rIII )
2H + 2H2 O.
(rIV )
O2 + 3H2
where we included reaction rII again. It should be noted that these overall reactions can be obtained
by the same algorithm which performs the linear algebra mentioned earlier (Appendix B of [15]).
Some aspects of the overall reactions are worth mentioning. First, we see that fuel-breakup rI
itself is chain-breaking since it consumes radicals. It is for this reason that the H-radicals are not able
81
to diuse upstream, as is observed in many applications. The H radical does not enter the fuel rich
zone in pre-mixed ames. Furthermore, it is interesting to see that not O2 but H2 O acts as the oxidizer
in the fuel breakup. Overall reaction rII is the water-gas shift-reaction, which accounts for the CO
to CO2 conversion in the post-ame zone. The recombination reaction rIII accounts for the chainbreaking in the post-ame zone. Its principal rate is given by reaction r5 (H + O2 + M
HO2 + M)
and apparently not r25 (H + H + M
H2 + M) as one might expect. Chain-branching is accounted for
by the oxygen breakup reaction, rIV , which eectively produces two radicals.
However, the set of equations is not complete yet. We may not expect that all steady-state species
have disappeared from the rates qI . . . qIV of the overall reactions. This would have been pure coincidence and indeed is not the case as one can see easily. Even if one would truncate the overall
rates to the principal rates (q11 , q9 , q5 and q1 , respectively) which is the minimum set, we still need
expressions for the concentration of HO2 , O and OH since these occur in reactions r5 , r1 and r9 . The
set (5.95.15) is not closed and we need expressions for most steady state species. These have to
be solved from the steady-state subset of equations in (5.8) leading to a set of nonlinear algebraic
equations, which is complicated to solve as it stands. In simulations this has to be done for each mesh
point, which is in general very CPU time consuming in 1D and specically in 2D and 3D. Therefore,
one often tries to truncate the expression for the steady-state species in such a way that an easy-toevaluate explicit set of equations evolves. The truncation procedure is again a sort of post-processing
procedure since the strategy for the truncation procedure can be obtained only after carefully examining the separate parts in the expressions for the steady state species which follow from a simulation
with the detailed chemical model. The procedure for lean methane-air ames [7] will be presented in
the next subsection.
5.3.4
Truncation of the steady-state relations
Formally for the evaluation of the concentrations of the steady-state species we should solve the set:
q10 + q11 + q12 + q13 q18 = 0
{CH3 }
q18 q19 q20 = 0
{CH3 O }
q13 q14 q15 + q19 + q20 = 0
{CH2 O }
q14 + q15 q16 q17 = 0

q1 q2 + q4 q13 + q18 = 0
q1 + q2 q3 2q4 + 2q6 q8 q9 q12 q15 + 2q22 q23 q24 = 0
q5 q6 q7 q8 2q21 + q23 = 0
q21 q22 q23 = 0
{HCO }
{O }
(5.17)
{OH }
{HO2 }
{H2 O2 }.
for xed non steady-state species and temperature, which is non-linear and coupled in the steady-state
species concentrations. To reduce the amount of work, which is one of the main objectives of the
reduction strategy, the complexity is to be reduced. Therefore truncation of this set has been given
much attention in the nineties and can be considered to be the most important step in reducing the
detailed chemical model from a numerical point of view. The approach which is used by Peters et al.
to truncate the steady-state expressions is presented.
Truncation thats what it is.
First we determine the minimum set of expressions we have to deduce in order to perform a ame
calculation with the reduced mechanism. Simple bookkeeping leads to the observation that in order to
82
r1
mole cm-3s-1
r4
Figure 5.3: The net reaction rate of OH for an adiabatic = 1.0 ame as a function of place. The
lines show the largest elementary reactions. Reactions r1 and r3 are the ones used in the truncated
expression. The listed numbers correspond to the
ones in table 2.1 on page 30 of lecture 2.
r12
r15
r3
Distance (cm)
calculate the set (qI )(qIV ) as presented earlier, we need expressions for all steady-state species. If the
overall rates would be restricted to the principal rates alone, this minimum set would be reduced to
the species CH3 , O, OH and HO2 . Note that, since the steady-state relations for these species involve
many reactions, there is no guarantee that the evaluation of the obtained expressions does not involve
the other steady-state species as well.
We now continue with the deduction of expressions as proposed by Peters [5]. The way in which
these expressions are obtained is by analyzing the source term of the dierent species. At rst, a
truncated expression for the concentration of OH is derived. For an estimate of the magnitudes of the
contributions of the dierent elementary reactions consider gure (5.3). In the reaction zone the main
contribution stems from reaction r3 which leads to a leading order approximation8 for OH:
[OH] =
r
k3 [H2 O][H]
f
k3 [H2 ]
(5.18)
The next extension can be obtained if the forward rate, qf , of reaction r1 is included, since this rate has
1
a non-negligible contribution in the reaction layer. However, if only qf would be added this leads to
1
an expression for [OH] which does not satisfy the correct equilibrium value downstream. Therefore,
the backward rate of r1 has to be added as well. The contribution of this backward rate is very small
and can be viewed upon as a second-order correction to the expression for the OH concentration.
The expression for the OH concentration now follows from balancing q1 and q3 :
f
r
f
r
[H2 ][OH]k3 [H2 O][H]k3 [H][O2 ]k1 + [O][OH]k1 = 0.
(5.19)
Since an explicit relation for [OH] is the principal goal, the yet unknown [O] in equation (5.19) should
be replaced by an expression in terms of known quantities. Once more, it should be kept in mind
that at this point the concentrations of the non steady-state species are assumed to be known and the
only other unknown is the O-concentration. The elimination is performed by introducing an ad hoc
approximation for the O-radical satisfying the partial equilibrium of reaction 4 leading to:
[O] =
f
[OH]2 k4
r
[H2 O]k4
r 2
f
[H2 O][H]2 k4 k3

.

=
f
r
[H2 ]2 k4 k3
8
In fact this leads to a partial equilibrium assumption of reaction 3.
(5.20)
83
Note that the leading-order approximation (5.18) for [OH] is substituted in the partial equilibrium
expression. Substituting this downstream value9 in the balance equation (5.19) leads to an explicit
expression for [OH]:
[OH] =
f
r
[H2 O][H]k3 + [H][O2 ]k1
r f
f
r f r
[H2 ]k3 + [H]2 [H2 O]k1 k4 k3 2 / [H2 ]2 k4 k3
(5.21)
This manipulation is expected not to be seriously in error, since the term involving the O radical (qr )
1
was only added to yield the correct equilibrium value for [OH] in the region downstream of the ame.
It should be stressed that the ad hoc expression (5.20) does not necessarily lead to a correct value
for [O] itself. In fact it is not correct as we can see when proceeding with a more systematic approach
to estimate the O-radical concentration. Following the same procedure for O as for OH the main
contributions are determined. This leads to the balance equation:
f
r
f
r
f
r
[H][O2 ]k1 + [OH][H]k2 + [OH]2 k4 = [O] k1 [OH] + k2 [H2 ] + k4 [H2 O] + k13 [CH3 ]
(5.22)
Now a problem arises because a non-negligible contribution (i.e. r13 ) necessitates the evaluation of an
expression for [CH3 ] which is not available yet. However, from the balance equation for CH3 ,
f
f
r
r
r
[H]k11 + [OH]k12 [CH4 ] = [CH3 ] k10 [H][M] + k11 [H2 ] + k12 [H2 O] + k13 [O]
(5.23)
it can be observed that reaction 13 links both equations. But, except OH, which is known at this
point, no other steady-state variables are involved. Therefore and due to the simple coupling between
equation (5.22) and equation (5.23), this set can be solved in closed form leading to a quadratic
expression for [O] yielding,
b + b2 4ac
[O] =
(5.24)
2a
where
a = k13 B,
b = BD + k13 (C A),
c = AD
(5.25)
and,
f
r
f
A = [H][O2 ]k1 + [OH][H]k2 + [OH]2 k4
r
f
r
B = [OH]k1 + [H2 ]k2 + [H2 O]k4
f
f
C = [H]k11 + [OH]k12 [CH4 ]
(5.26)
r
r
r
D = [H][M]k10 + [H2 ]k11 + [H2 O]k12 .
Using the expression for the O-radical gives,

[CH3 ] =
C
D + k13 [O]
(5.27)
which is the last species needed for the principal rates. Note that already an evaluation order for the
steady-state relations appears, rst OH and then O and CH3 have to be calculated.
9
Note that this expression may only be expected to be a correct estimate for [O] downstream, where the mixture approaches equilibrium.
84
Expressions for the other species can be derived as well. Once the expressions for OH, O, and
CH3 are known, the truncated values for the other steady state species are easy to obtain. Below the
results for the remaining intermediates CH3 O, CH2 O and HCO in the main chain are listed,
[CH3 ][O2 ]k18
[H]k19 + [M]k20
[CH3 ][O]k13 + [CH3 O] ([H]k19 + [M]k20 )
[CH2 O] =
[H]k14 + [OH]k15
([H]k14 + [OH]k15 ) [CH2 O]
[HCO] =
[H]k16 + [M]k17
[CH3 O] =
(5.28)
(5.29)
(5.30)
The expressions for ([HO2 ]) and hydrogen-peroxide ([H2 O2 ]) can be obtained from their main rates
in the H2 -O2 subsystem (r5 - r8 ). This leads to the following expressions,
[HO2 ] =
[H2 O2 ] =
[H][O2 ][M]k5
[H]( k6 + k7 ) + [OH]k8
r
r
[HO2 ]2 k21 + [OH]2 [M]k22 + [H2 O][HO2 ]k23
f
f
[M]k22 + [OH]k23
(5.31)
(5.32)
It should be stressed that the above set of equations form explicit expressions if they are evaluated in
the order (5.21)(5.32). The steady-state expressions together with the dierential equations for the
non steady-state species and the chemical source terms given in section (5.3.3) dene the complete
system.
5.4
Mathematical systematic reduction techniques
The major drawback of the conventional approach towards the systematic reduction is the need
for experience and deep insight into reaction mechanisms. Furthermore, once the mechanism is selected it will be used throughout the whole domain of the composition space where the particular
application resides. It might well be true that the used mechanism is excellent in a certain temperature/composition range but completely wrong for the rest. The mathematical reduction techniques
try to circumvent these drawbacks by applying rigourous mathematical methods which decide locally
what appropriate steady-state candidates are. Here the mathematical framework will be sketched.
However details on implementation can be found in e.g. [16, 17].
For the analysis it is sucient to consider the balance equations for species (equation 5.6). To
study the chemical dynamics the convective and diusive uxes are neglected, yielding
i
= wi
t
(5.33)
where wi = i j q j . To be able to determine mathematically the fast and slow processes around a certain point in composition space 0 = (0 , . . . , 0 )T , the source term w = (w1 , . . . , wN )T is linearized
N
1
around this reference:
w w(0 ) + J( 0 )
(5.34)
the J = w
the analyis:
, denotes the Jacobian matrix. Substitution of equation 5.33 gives the starting point of
w(0 ) + J( 0 ).
t
(5.35)
85
The local characteristics of the system can be examined by transforming these equations into the
basis of the eigenvectors of the Jacobian matrix. The eigenvectors are si are dened by:
J si = i si
with i the eigenvalue of the Jacobian matrix. If the transformation matrix is dened according to,
S = (s1 , . . . , sN ), the former equation writes,
JS = S
(5.36)
where10 = diag(1 , . . . , N ) is a matrix with only non-zero elements on its diagonal. Multiplication
of equation 5.36 on the left by S1 yields:
= S1 JS.
(5.37)
It should be noted the the eigenvalues and eigenvectors may be complex at this point. We will not
go into details but in a practical implementation of this method this should be tackled. Dierences in
implementation found in literature depend on how a new basis is formed from the eigenvectors. Some
construct an orthogonal basis (Schur-basis [16]), others use the eigenvectors as is (e.g. [17]).
2
1.8
1.6
REAL i
(i) > 0
i >>1
1.4
0.5
1.2
Equation 39
Equation 39
COMPLEX i
1.5
0.8
0.5
i < 1
0.6
(i) < 0
1
0.4
0+i*t
0.2
0.05
0.1
0.15
0.2
0.25
0.3
0.35
1.5
0.4
0.05
time
0.1
0.15
0.2
0.25
0.3
0.35
0.4
time
Figure 5.4: On the left: Equation 5.39 for i >> 1 and < 1. The asymptotic solution i = 0 + wi t is given
i
with the dotted line with circles. On the right: Equation 5.39 for |i | >> 1 with (i ) > 0 and (i ) < 0
respectively.
Now we can use the eigenvector basis to transform equation 5.35 according to = S thus
yielding:
0
0

w( ) + ( ).
(5.38)
t
Whats the whole point of this exercition? The orginal dierential equations in equation 5.35 are now
transformed such that the individual equations
wi (0 ) + i (i 0 )
i
i
t
10
And NOT JS = S!
86
are independent from each other, they are decoupled. For each one the general solution is given by:
i = 0 +
i
wi (0 ) t
i
e i 1
i
(5.39)
This equation gives now the basis for a timescale analysis of dierent chemical processes. These
processes, in general, can not be associated with a single species or reaction, but inevitably consider
reaction groups. From equation 5.39 typical timescales can be distinguished. If the absolute value
obeys |i | 1 equation 5.39 becomes11
i = 0 + wi t
i
indicating that typical timescales are
i = 1/wi
|i |
1.
(5.40)
If the absolute value of the real part of an eigenvalue | (i )| 1, the typical timescale is found from
the exponent:
i = 1/| (i )| | (i )| 1.
(5.41)
Several observations follow from the value of i . If its magnitude is large, movements in the
direction of the associated eigenvector will proceed fast. The larger the eigenvalue the faster the
process will be. Dierent ways of behaviour are determined by the sign of the real part and the
magnitude of the imaginary part. A process with a negative real part of the eigenvalue will extinguish
and reach a steady-state. If the imaginary part is large as well this relaxation will be oscillatory. If
the real part of the eigenvalue is positive the associated process will not relax to a steady-state. This
is illustrated in gure 5.4. Some eigenvalues will be zero. One can show rigourously that the fact
that chemical reactions conserve elements can be used to construct eigenvectors that have eigenvalues
equal to zero. In fact all eigenvectors with zero eigenvalue are associated with so-called conserved
scalars.
Similar to the conventional approach now the mathematical method will introduce steady-state
relations, albeit generally not coupled to specic species. The best candidates are selected from the
ordering of the eigenvalues. The fastest timescales pose the best candidates. The number of candidates
to be selected is determined by the dimension12 of the so-called Low Dimensional Manifold to
be created. In order to be successful it is essential that many eigenvalues have negative real parts
while their magnitude is large. That this is the case has been shown by e.g. Maas and Pope [16]
for carbonmonoxide/hydrogen-air combustion systems, and in the PhD Thesis of Eggels [17] for
hydrogen-air en methane-air systems.
The number of steady-state relations, or equally, the number of fast reaction-groups, n s to be
selected is given by the required dimension, say nm , of the manifold together with the number of
conserved scalars (viz. elements) nc . The so-called elemental composition vectors i , with elements
i j the number of elements i in species j are left eigenvectors of the Jacobian with eigenvalue 0.
Therefore, the number of steady-state assumptions n s within the the set of i obeying a steady-state
equation,
i
= 0 i = nm + nc + 1, , n
(5.42)
t
11
Use lim e x = 1 + x
x0
12
The dimension is determined by the degrees of freedom in composition space, being the number of species n. The
realizable compositions will be restricted by imposing steady-state relations generating a hyper-surface in composition
space. This surface is mostly referred to as the Low-dimensional Manifold
87
only equals n s = n (nm + nc + 1). Here n is the total number of species present in the model or
equivalently the dimension of the composition space.
References
[1] Smooke, M.D., Xu, Y., Zurn, R.M., Lin, P., Frank, J.H. and Long, M.B., Computational and Experimental Study of OH and CH Radicals in Axisymmetric Laminar Diusion Flames in TwentyFourth symposium on Combustion The Combustion Institute, Pittsburgh, p.813, 1992
[2] H.C. de Lange, Modeling of premixed ames, Ph.D. Thesis, University of Technology, Eindhoven 1992.
[3] van Maaren, A., One-step chemical reaction parameters for premixed laminar ames, Ph.D.
Thesis, Eindhoven University of Technology, 1994.
[4] C.K. Westbrook and F.L. Dryer, Simplied Reaction Mechanisms for the Oxidation of Hydrocarbon Fuels in Flames., Combustion Science and Technology 27, p. 31, 1981
[5] Peters, N., in Reduced Kinetic Mechanisms for Asymptotic Approximations for Methane-Air
Flames (Smooke, M.D., Ed.), Lecture Notes in Physics 384, Springer-Verlag, Berlin, p. 48,
1991
[6] Smooke, M.D. ed., Reduced Kinetic Mechanisms and Asymptotic Approximations for MethaneAir Flames, Lecture Notes in Physics, Springer-Verlag, Berlin Heidelberg, 1991
[7] Mauss, F.,Peters, N., in Reduced Kinetic Mechanisms for Applications in Combustion Systems
(Peters,N., Rogg,B., Eds), Lecture Notes in Physics m15, Springer-Verlag, Berlin, p. 58, 1993
[8] Bowman,G., Frenklach,M., Gardiner, B., Smith,G. and Serauskas,B, GRI-Mech, Gas Research
Institute, Chicago, Illinois, USA, http://www.me.berkeley.edu/gri-mech/index.html
[9] Bilger,R.W. Strner, S.H. and Kee,R.J., On reduced mechanisms for Methane-Air Combustion
in Nonpremixed Flames, Combustion and Flame 80, p. 135, (1990)
[10] Seshadri, K. and Peters, N., The inner structure of Methane-air ames., Combustion and Flame
81, p. 96, 1990
[11] Peters, N. and Williams, F.A., The asymptotic structure of Stoichiometric Methane-air ames.,
Combustion and Flame 68, p. 185, 1987
[12] Paczko G., Lefdal, P.M. and Peters, N., Reduced reaction schemes for methane, methanol amd
propane ames, Twenty-rst Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, p. 739, 1986
[13] Boersma, John , Modelling of NO x emission from natural gas red turbine combustors, Ph.D.
Thesis, University of Technology, Twente 1993.
[14] G ttgens,J., Terhoeven, P., in Reduced Kinetic Mechanisms for Applications in Combustion Syso
tems (Peters,N., Rogg,B., Eds), Springer-Verlag, Berlin, p. 343, 1993
Ph.D. Thesis, Eindhoven University of Technology, 1994
88
[16] Maas, U and Pope, S.B., Simplifying Chemical Kinetics: Intrinsic Low-Dimensianal Manifolds
in Composition Space, Combustion and Flame 88, p. 239, 1992
[17] Eggels, R.L.G.M., Modelling of Combustion Processes and NO formation with Reduced Reaction Mechanisms, Ph.D. Thesis, Eindhoven University of Technology, 1996
Chapter 6
Flamelet-Generated Manifolds
J.A. van Oijen
6.1
Introduction
Although the speed and storage capacity of modern computers increases continuously, it is still impossible to employ models that use detailed chemistry for the simulation of combustion in practical
furnaces. Especially the models for turbulent ames are tremendously complicated due to the large
range of time and length scales involved. The modelling of laminar ames is easier because the ow is
relatively simple. Nowadays, numerical simulations of one-dimensional laminar ames with detailed
chemistry and transport models are often used in combustion research. Nevertheless, the computation time for multi-dimensional laminar ame models prohibits the simulation of ames in complex
geometries or an investigation of the inuence of dierent parameters.
The use of complex reaction mechanisms is, apart from some simple cases, impossible, because
for each species a partial dierential equation has to be solved, and a typical reaction mechanism for
hydrocarbon combustion considers more than 100 species [32]. Furthermore, the chemical kinetics
cause the system of dierential equations to be very sti, which makes the use of expensive implicit
solvers inevitable. Therefore, much research is spent on the development of reduction methods, which
create reduced chemical models based on the complex reaction mechanism. These simplied chemistry models reduce the computational eort considerably without losing too much accuracy.
The rst and probably the most well-known systematic reduction method is introduced by Peters et
al. [21, 22]. This method (referred to as conventional reduction technique) is based on the observation
that due to very fast chemical processes in combustion problems many chemical species and reactions
are in quasi-steady state or partial equilibrium. When a species is assumed to be in steady state,
the corresponding dierential equation can be replaced by an algebraic relation, which reduces the
computational cost. The main disadvantage of the method is that a vast knowledge of the chemical
kinetics is required, because the researcher has to decide which species can be assumed to be in steady
state. This problem is solved in the reduction methods introduced by Lam and Goussis [12] and Maas
and Pope [13]. In these reduction techniques a more mathematical approach is used, which makes it
possible to perform the reduction procedure automatically without insight of the user in the chemical
kinetics. In both methods the fast chemical processes are identied by a mathematical analysis of the
chemical system. Linear combinations of species corresponding to the fast chemical processes are
assumed to be in steady state. In the method of Maas and Pope the fast and slow chemical processes
are separated by using an eigenvalue analysis of the Jacobian of the chemical source term, while Lam
and Goussis propose a computational singular perturbation method.
The methods based on steady-state assumptions are, however, less ecient in low-temperature
regions. At high temperatures many chemical processes are fast and can be assumed to be in partial
89
90
Chapter 6 J.A. van Oijen
equilibrium. At lower temperatures the number of slow chemical processes increases and less species
can be assumed to be in quasi-steady state. At these lower temperatures, convection and diusion
become important and the local mixture composition is determined by a balance between convection,
diusion and reaction.
In this lecture a reduction method is discussed, that is not only based on chemical assumptions
like existing reduction methods, but that also takes the main parts of convection and diusion into
account. It shares the idea with so-called amelet approaches [11, 19] that a multi-dimensional ame
can be considered as a set of one-dimensional (1D) ames: amelets. This implies that the path
followed in composition space in case of multi-dimensional ames will be close to the path found in
1D ames. Therefore, the chemical compositions in amelets are used to construct a low-dimensional
manifold in this method. The resulting manifold is referred to as a Flamelet-Generated Manifold
(FGM).
In the following section a set of amelet equations is derived, starting from the full set of transport
equations. Then it is demonstrated how the solutions of the set of amelet equations can be used to
construct a manifold. The application of FGM in combustion simulations is discussed as well. Finally,
some test results for 1D and 2D burner-stabilized methane/air ames are presented.
6.2
Flamelet equations
The FGM method shares its basic assumption with amelet methods, that a multi-dimensional ame
can be considered as a set of 1D amelets. Therefore, the compositions in 1D ames will be representative for the compositions in more general ames as well. In this section a set of 1D amelet
equations is derived following the basic ideas of De Goey and Ten Thije Boonkkamp [10, 11]. Although this derivation has been presented in a previous lecture, it will be repeated here for the sake of
completeness. In the following sections it is explained how the set of amelet equations can be used
to construct a FGM.
A premixed ame is dened as the region in space, where a scalar variable Y assumes values
between Yu = 0 in the unburnt gases and Yb = 1 in the burnt gases. The subscripts u and b denote
variables associated with the unburnt and burnt mixture, respectively. The variable Y can be any
linear combination of species mass fractions, which obeys Y 0. A ame surface is dened as an
iso-surface of Y, i.e. a surface at which Y(x, t) = const. The motion of such a surface is described by
the kinematic equation
dY
Y
:=
+ u f Y = 0,
(6.1)
dt
t
stating that a point on a ame surface stays on this surface for all t. The local velocity of a ame
surface u f is determined by a balance between the local uid velocity u and the local burning velocity
sL :
u f = u + sL n.
(6.2)
The burning velocity is dened as the velocity at which the ame surface propagates normal to itself
and relative to the ow in the unburnt mixture. The local normal vector n can be determined from the
scalar eld Y,
Y
,
(6.3)
n=
| Y|
directed to the unburnt gas mixture. As gure 6.1 shows, the tangential component of the ame surface
velocity is equal to the tangential component of the uid velocity and the dierence (u u f ) normal
91
u
Y(x, t) = Y3
Y(x, t) = Y2
uf
Y(x, t) = Y1
sL n
Figure 6.1: Flamelet system adapted to iso-surfaces of Y.
to the ame surface is equal to the amount of mass eectively consumed by the ame. Substitution of
(6.2) in (6.1) leads to the kinematic equation
Y
+ u Y = sL | Y|.
t
(6.4)
This equation is very similar to the well-known kinematic G-equation rst introduced by Williams
[34], which describes the motion of a single ame sheet (Y = Y0 ). However, important dierences
are that (6.4) describes the motion of all ame surfaces with Yu < Y < Yb , and that sL is a eld
quantity in (6.4), while in the G-equation it is dened at the ame sheet only.
In [10], De Goey and Ten Thije Boonkkamp proposed the stretch rate K as the relative rate of
change of the mass M(t), dened as
M(t) =
dV,
(6.5)
V(t)
contained in an innitesimal volume V(t) in the ame, moving with velocity u f :

K=
1 dM
.
M dt
(6.6)
Applying Reynolds transport theorem to M(t) in (6.5) gives the following expression for the scalar
eld quantity K:
K =
+ (u f ).
(6.7)
t
Together with (6.2) this expression can be used to rewrite the continuity equation:
(sL n) = K,
(6.8)
where all distortions from local 1D ame behaviour are combined in the right-hand side term K.
Apart from the usual terms related to ame curvature and ow straining, this denition of K incorporates additional contributions, for instance due to ame thickness variations. A physical interpretation
of the dierent contributions to ame stretch is presented in [10].
The conservation equation for the scalar variable Y reads
(Y)
+ (uY)
Y Y = 0,
t
LeY c p
(6.9)
92
with LeY and Y generally depending on the other eld variables in the ame. Using the kinematic
equation (6.4) and equations (6.7) and (6.2), the following equation for Y is found rigorously
(sL nY)
Y Y = KY,
LeY c p
(6.10)
with all transport terms along the ame surfaces gathered in the right-hand-side term KY. Introducing the mass burning rate m = sL , the arc-length perpendicular to the ame surfaces s and
the variable , which is a measure for the ame surface through which transport takes place (cf.
gure 6.1), equation (6.10) can be written in quasi-1D form normal to the ame surfaces:
Y
(mY)
Y = KY.
s
s LeY c p s
(6.11)
The derivative of is related to the curvature of the ame surface via

= n =
1
.
s
(6.12)
Using this expression for , (6.11) becomes

Y
Y = KY + FY ,
(mY)
s
s LeY c p s
(6.13)
where all perturbations from 1D at ame behaviour are gathered at the right-hand side. In (6.13)
we introduced FY as a short-hand notation for the convective and diusive ux of Y. Using similar
notations for the uxes of the species mass fractions and the enthalpy, i.e.
Fi = mYi
and
Fh = mh
c p s c p
Yi
,
Lei c p s
Ns
i=1
(6.14)
1
Yi
1 hi
,
Lei
s
(6.15)
the full set of conservation equations can be cast in a quasi-1D form, which we refer to as the amelet
equations:
m
= K + m,
s
(6.16)
Fi
i = KYi + Fi + Qi ,
s
Fh
= Kh + Fh + Qh .
s
(6.17)
(6.18)
This set of equations describes the internal structure of the ame front in terms of Yi (s), hi (s) and the
mass burning rate m(s) for a amelet with a particular stretch eld K(s) and curvature eld (s). When
the set (6.16)(6.18) is solved in combination with the kinematic equation (6.1) for all ame surfaces,
the full set of conservation equations for species and enthalpy is solved. The terms Qi and Qh describe
transport along the ame surfaces, which arises because the local iso-surfaces of the variables Yi and
h generally do not coincide with iso-surfaces of Y. The terms Qi and Qh are given by
Qi = (u f i u f ) Yi +
Lei c p
|| Yi
(6.19)
93
ufi
uf
(u f i u f ) Yi
Lei c p
|| Yi
Figure 6.2: Schematic representation of the dierent contributions in Qi . The black and gray lines denote
iso-surfaces of Y and Yi , respectively.
and
Qh = (u f h u f ) h +
cp
|| h +
cp
Ns
i=1
1
1 hi
Lei
Yi ,
||
(6.20)
with u f i and u f h the local velocities of the iso-surfaces of Yi and h, respectively. In (6.19) and (6.20),
|| denotes the nabla operator in tangential direction only. As shown in gure 6.2, the rst term on the
right-hand side of (6.19) represents the relative movement of the iso-surfaces of Yi with respect to the
iso-surfaces of Y. The second term denotes the diusive transport of Yi along the iso-surfaces of Y.
So far, no approximations have been made. The set of amelet equations (6.16)(6.18) are found
rigorously from the full set of conservation equations by using the stretch rate equation (6.7) and
the kinematic equation (6.4). All perturbations from local 1D at ame behavior are gathered in the
right-hand sides of the amelet equations. It is expected that in most situations in premixed laminar
ames these perturbations are small compared to the other terms in the amelet equations, although
this might not be justied under extreme circumstances, such as near local ame quenching. The
stretch terms in (6.16)(6.18) can be neglected when the Karlovitz number is small:
Ka =
Kf
sL
1,
(6.21)
with f the ame thickness. Curvature eects are negligible when the curvature radius 1 of the ame
surfaces is much larger than the ame thickness:
|1 |
f .
(6.22)
If the transient time scales are longer than the ame time scale f /sL , the rst term in Q can be
neglected as well. The second term in Q was computed numerically for the tip of a 2D Bunsen ame
in [9] and it appeared to be very small. The approximation that this term is negligible, is related to the
assumption that the length scales of the distortions along the ame surfaces are much larger than the
ame thickness.
94
If all perturbations from 1D at-ame behaviour are neglected, the following set of 1D equations
remains:
m
= 0,
s
Yi
(mYi )
= i ,
s
s Lei c p s
(mh)
h

s
s c p s
s c p
(6.23)
i = 1, . . . , Ns ,
Ns
i=1
Yi
1
1 hi
Lei
s
(6.24)
(6.25)
The balance equation (6.24) between convection, diusion and reaction can be considered as a steadystate relation. Note that in the steady-state relations of conventional reduction techniques not only the
multi-dimensional perturbations are neglected, but also the left-hand side of (6.24) describing the
major parts of convection and diusion.
The set of 1D equations (6.23)(6.25) can be solved relatively easily with specialized 1D ame
codes like CHEM1D [3]. A solution of this set of equations is called a amelet and forms a curve in
composition space, which can be regarded as a one-dimensional manifold.
6.3
Manifold construction
In this section it is described how the set of 1D equations (6.23)(6.25) is used to create a manifold. As
in the ILDM method a distinction is made between variables that are changed by chemical reactions
(the species mass fractions Yi ) and variables that are conserved by chemical reactions (Z j , p and h).
The composition of the chemical equilibrium mixture is completely determined by the values of the
conserved variables, while the combustion process from the unburnt to burnt state is described by the
reactive controlling variables, which are referred to as progress variables. Following Maas and Pope
[13], we rst consider the case of constant conserved variables. In section 6.3.3 it is discussed how
variations in the conserved variables should be treated.
6.3.1
One-dimensional manifold
In order to construct a FGM, the set of 1D amelet equations (6.23)(6.25) is solved treating the
system as an adiabatic, freely-propagating, premixed, at ame. This implies that the boundary conditions at the unburnt side are of Dirichlet type
Yi (s ) = Yi, ,
h(s ) = h ,
(6.26)
dh
(s ) = 0,
ds
(6.27)
and at the burnt side of Neumann type

dYi
(s ) = 0,
ds
and that the mass burning rate m is an eigenvalue of the problem. The solution of this system is a 1D
curve in composition space and is determined by its starting point (Yi, , h ). The one-dimensional
FGM in composition space is simply the amelet starting at the point that represents the unburnt mixture for which the manifold is created. In this way the manifold connects the two most distinguished
points in composition space: the point corresponding to the unburnt mixture (Yi, , h ) = (Yi,u , hu )
and the equilibrium point (Yi, , h ) = (Yi,eq , heq ).
Application of Flamelet-Generated Manifolds
95
93
T (K)
400
800
1200
1500 2000 2400
4.0
3.5
3.0
YH 103
2.5
2.0
1.5
1.0
0.5
0.0
0
0.05
0.1
0.15
YH2 O
0.2
0.25
Figure 7.3: Projection of the 1D FGM (full line) and ILDM (dashed line) onto the 2 OH plane for a stoichioFigure 6.3: Projection of the 1D FGM (full line) and ILDM (dashed line) onto the HH2 OH plane for a stoichiometric hydrogen/air mixture. The gray dots represent the chemical state at different positions in a 2D premixed
metric hydrogen/air mixture. The gray dots represent the chemical state at dierent positions in a 2D premixed
hydrogen/air ame computed with detailed chemistry. The temperatures indicated the top axis correspond to
hydrogen/air ame computed with detailed chemistry. The temperatures indicated atat the top axis correspond to
the FGM.
the FGM.
As an example the one-dimensional FGM and ILDM for a stoichiometric hydrogen/air mixture
FGM in composition space is simply the amelet starting at the point that represents the unburnt mixare shown in gure 6.3 together with a scatter plot of the chemical state at dierent positions in a 2D
ture for which the manifold is created. In this way the manifold connects the two most distinguished
stationary premixed Bunsen ame computed with detailed chemical kinetics. The detailed reaction
points in composition space: the point corresponding to the unburnt mixture (Yi, , h ) = (Yi,u , h )
mechanism is a subset of the reaction mechanism developed by Smooke [27] and includes 7 speciesu
and the equilibrium point (Yi, , h ) = (Yi,eq , h eq ).
and 7 reversible reactions. The mass fraction of atomic hydrogen YH is plotted as function of the mass
As an example the one-dimensional FGM and ILDM for a stoichiometric hydrogen/air mixture
fraction of water YH2 O , which is a suitable progress variable in this case, because it is continuously
are shown in gure 7.3 together with a scatter plot of the chemical state at different positions in a 2D
increasing in the ame from YH2 O,u = 0 in the unburnt mixture to the chemical equilibrium value
stationary premixed Bunsen ame computed with detailed chemical kinetics. The detailed reaction
YH2 O,eq = 0.24. If YH had been used as controlling variable, it would not have led to a unique mapping.
mechanism is a subset of the reaction mechanism developed by Smooke [27] and includes 7 species
Generally, Lewis-number (or dierential diusion) eects cause local enthalpy and element mass
and 7 reversible reactions. The mass fraction of atomic hydrogen YH is plotted as function of the mass
fraction variations in ames andis a suitable progress variable in In the ILDM method,ishowever, the
fraction of water YH2 O , which thus also in the FGM database. this case, because it continuously
element composition and from Y
of 0 in the unburnt mixture to the chemical equilibrium value
increasing in the ame the enthalpy = the mixture are constant throughout the manifold and variH2 O,u
ations in the conservedYvariables canused as controllingfor by using additional controllingto a unique
only be accounted variable, it would not have led variables.
YH2 O,eq = 0.24. If H had been
Therefore, to make a fair comparison of the methods in gure 6.3, Lewis numbers equal to 1 have
mapping.
been Generally, Lewis-numberof both the FGM and the 2D ame so that dierential and element mass
used in the computation (or differential diffusion) effects cause local enthalpy diusion eects
are absentvariations in cases. An example ofin the FGM database. InLewis numbers unequal to 1, is
fraction in all three ames and thus also a FGM computed with the ILDM method, however, the
presented composition and the enthalpy of the mixture are constant throughout the manifold and varielement in section 6.3.3.
Because conserved variables can only be accounted processes additional temperature range
ations in thethe composition is determined by chemicalfor by usingin the highcontrolling variables.
(YH2 O > 0.19, make 1600 K), the ILDM and FGM are in gure 7.3, Lewis numbers in the to 1 have
Therefore, to T > a fair comparison of the methods there equivalent. However, equal colder
zone (YH2 O in the computationprocesses disturb the balance ame so that differential diffusion effects
been used < 0.19) diusion of both the FGM and the 2D between chemical production and consumption. Since the major parts of example of and diusion are included in FGM, the FGM lies to 1, is
are absent in all three cases. An convection a FGM computed with Lewis numbers unequal much
closer to the data of the detailed chemistry computation than the ILDM. In order to get accurate represented in section 3.3.
96
sults in this region, the ILDM should be extended with additional progress variables, which decreases
the numerical eciency of the method. For this simple hydrogen reaction mechanism two progress
variable will probably be enough. For more complex fuels and mechanisms many more controlling
variables are needed to create an accurate ILDM. The addition of extra dimensions to the ILDM in
order to satisfy accuracy requirements is described by Schmidt et al. [25].
The 2D Bunsen ame plotted in gure 6.3 is stabilized on a cold burner rim and is therefore locally
cooled. This results in lower H radical concentrations than for the adiabatic FGM. In section 6.3.3 it
is explained how these non-adiabatic eects can be accounted for in FGM.
6.3.2
Multi-dimensional manifolds
Like in the ILDM method, the accuracy of the FGM method can be enhanced by increasing the
dimension of the manifold. With each additional dimension an extra progress variable, as well as
the corresponding dierential equation, is introduced. In this way an extra degree of freedom is
added to the reduced system, which results in a more accurate description of the full system. When
the dimension of the manifold is increased, a larger part of the composition space is covered. The
addition of dimensions can be continued until the manifold spans the complete composition space. In
that case the full system is recovered.
A 2D FGM can be constructed from a set of amelets starting at dierent points on a 1D curve
in composition space with constant conserved variables. This set of 1D amelets forms a 2D surface
in composition space: the 2D FGM, which can be parametrized by Npv = 2 progress variables. This
method to generate multi-dimensional manifolds has been introduced by Pope and Maas [24]. A
unique choice of the starting curve manifold generator in [24] cannot be given, although it
should include the unburnt mixture and must be chosen in such a way that the resulting manifold
covers an as large as possible range of the progress variables. Following these arguments Pope and
Maas proposed a specication of the starting curve based on the extreme values of major species [24].
For a methane/air system a sensible choice of the major species is: CH4 , O2 , CO, H2 , CO2 , H2 O and
N2 . A mixture of these major species is stable and does not react at room temperature (i = 0). The
other (or minor) species are taken zero at the starting curve.
The exact choice of the starting curve determines the resulting manifold, but it may not be crucial.
If we assume that two controlling variables are sucient to represent the chemical composition in a
premixed ame satisfactorily, then the amelets will rapidly relax towards a 2D attracting manifold.
Therefore, however the starting points are chosen, the resulting manifold will be close to the underlying manifold except near the starting curve itself, where chemistry is negligible anyhow. This method
to generate multi-dimensional manifolds can be extended to the general case of Npv dimensions by
choosing a (Npv 1)-dimensional starting plane.
As an example a 2D FGM is computed for a stoichiometric methane/air mixture. The projection
of this manifold on the CO2 CO plane is shown in gure 6.4. The origin of the plot corresponds to
the unburnt mixture, which consists of CH4 , O2 and N2 only, and it is connected with the equilibrium
point by the 1D manifold. The projection of the starting curve coincides with the horizontal and
vertical axis in gure 6.4. Along the horizontal axis CO2 and H2 O are added to the initial mixture,
while along the vertical axis CO and H2 are added. Along the whole starting curve the enthalpy and
the element mass fractions are kept constant by changing the values of YCH4 , YO2 and T , which causes
all amelets to end up in the same chemical equilibrium. The dierent amelets rapidly converge to
a 1D manifold, which indicates that, apart from the conserved variables, one progress variable might
be enough to describe the chemical state in that region.

95
97
YCO 10
0.8
0.6
0.4
0.2
0
0
0.2
0.4
0.6
0.8
YCO2 10
1.2
1.4
Figure 7.4: Projection of two-dimensional FGM onto the CO CO plane for stoichiometric methane/air
Figure 6.4: Projection of aatwo-dimensional FGM onto the CO22 COplane for aastoichiometric methane/air
mixture. The bold line denotes the one-dimensional manifold.
mixture. The bold line denotes the one-dimensional manifold.
theFor application in ame simulation codes, a manifold is values of YCH4 ,byOcontrolling variables.
element mass fractions are kept constant by changing the parametrized Y 2 and T , which causes
all ameletsmass fractions orsame chemical equilibrium. The different amelets rapidly convergeIn
The species to end up in the linear combinations of them can be used as controlling variables. to
a 1D and FGM the selection of the apart from (or conserved variables, not related variable slow
ILDMmanifold, which indicates that, controlling the progress) variables isone progressto fast or might
be enough to describe the chemical state reduction method. In fact any combination of species can
chemical processes as in the conventionalin that region.
For application in ame simulation the manifold leads parametrized by controlling mixture
be used as long as the parametrization ofcodes, a manifold isto a unique description of thevariables.
The species by thefractions or variables. For the 1D manifold shown in gure 6.4 withvariables. In
composition mass controlling linear combinations of them can be used as controlling a bold line,
ILDM and FGM the controlling the controlling (or would result in a non-unique mapping. or the
YCO2 is an appropriateselection of variable, while YCOprogress) variables is not related to fast Onslow
chemical is YCO a as in the conventional reduction method. 2D manifold.
other handprocessessuitable second controlling variable for the In fact any combination of species can
be used as long as the parametrization of the manifold leads to a unique description of the mixture
composition by the controlling variables. For the 1D manifold shown in gure 7.4 with a bold line,
6.3.32 isConserved variables
YCO
an appropriate controlling variable, while YCO would result in a non-unique mapping. On the
other hand is YCO a suitable second controlling variable for the 2D manifold.
In general, one or more of the conserved quantities p, h and Z j might vary in space or time due to
other processes than chemical reaction, e.g. mixing and cooling/heating. If variations in a conserved
3.3 Conserved variables
variable are expected to be important in the application, then this variable should be added to the
manifold as an additional of the conserved quantities p, h and Z j might vary inthe enthalpy changes
In general, one or more controlling variable. In burner-stabilized ames, e.g., space or time due to
due to processes thaneects resulting from ame stabilization. Furthermore, variations in to conserved
other non-adiabatic chemical reaction, e.g. mixing and cooling/heating. If cooling due a radiation
or walls may expected to be important in the application, thenthese variable should be the FGM the
variable are cause the enthalpy to decrease. To account for this enthalpy changes, added to is
extended with h additionaladditional parameter. In burner-stabilized ames, e.g., the enthalpyamelet
manifold as an being an controlling variable. The manifold is extended by solving the 1D changes
equations (6.23)(6.25) for dierent values of hstabilization. Furthermore, -dimensionalto radiation
due to non-adiabatic effects resulting from ame . Corresponding to the Npv cooling due manifold,
we then obtaincausepv + enthalpy to decrease. To parametrizedthese enthalpy changes, theand one
or walls may a (N the 1)-dimensional manifold account for by Npv progress variables FGM is
additional with h being an additional parameter. The manifold is extended by solving the 1D amelet
extended controlling variable h.
equations (7.23)(7.25) for different variables h be treated similarly, yielding a Ncv -dimensional
Changes in the other conserved values of can . Corresponding to the Npv -dimensional manifold,
we then obtain a (NpvNpv 1)-dimensional manifold of controlling by Npv progress av the number one
manifold with Ncv = + + Nav the total number parametrized variables and N variables and of
additional controlling variable N
additional controlling variables.h. av has a maximum value of Ne + 1, i.e. the number of independent
Changes in the However, only small number of additional controlling variables is needed for
conserved variables. other conserveda variables can be treated similarly, yielding a Ncv -dimensional
manifold with Ncvbecause in manythe total number of controlling variables and Nav the number In
most applications, = Npv + Nav cases several conserved variables can be assumed constant. of
98
open systems, for instance, the pressure is constant and does not have to be used as an additional
controlling variable.
In the remaining part of this chapter we discuss the application of FGM to laminar premixed
burner-stabilized ames. Since these ames are purely premixed, the element mass fractions are constant apart from local changes due to dierential diusion eects, which are included in the manifold.
Because the pressure is also constant in these ames, the enthalpy is the only conserved variable in
which signicant changes are expected (Nav = 1). Since no large multi-dimensional perturbations are
expected as well, one progress variable is used (Npv = 1), resulting in a 2D manifold. As explained in
section 6.3.2 more progress variables can be used to increase the accuracy of the method. Since this
implies that more dierential equations have to be solved and the look-up procedure in the database
becomes more complicated, the gain in computation time will decrease. Therefore, a small number of
controlling variables is preferred.
6.4
Application of FGM
The application of FGM in ame simulations can be divided in two parts. The rst part is the preprocessing step: the generation and storage of the amelet data. This is explained in section 6.4.1.
The second part, described in section 6.4.2, considers the equations to be solved during the ame
computation itself.
6.4.1
Storage and retrieval
The implementation of FGM in a reacting-ow code is similar to the implementation of ILDM. The
low-dimensional manifold is found by solving steady-state equations in a pre-processing step and the
manifold is stored in a look-up table. In the FGM method a manifold is constructed by using solutions
of the 1D amelet equations (6.23)(6.25).
To construct a 2D manifold with enthalpy variations, a series of amelets is computed for different values of h as mentioned in section 6.3.3. The rst amelet is solved for an initial mixture
with a relatively high temperature, say T ,max = 390 K. For the following amelets the enthalpy is
decreased by lowering the temperature of the initial mixture in steps of approximately 30 K. The 1D
amelet equations are solved by using a modied Newton method, in which the previous amelet is
used as the initial guess of the new solution. Since the dierence between two succeeding amelet
solutions is small, a few iterations are sucient to reach a converged solution. When an initial temperature of about T ,min = 240 K is reached, another method to lower h is applied, because lower
temperatures are not realistic. From hereon the enthalpy is decreased by converting a fraction of the
initial fuel with oxygen into the products CO2 and H2 O, keeping the element composition constant and
the initial temperature at T ,min . This process is continued until the ame temperature T b becomes
too low and the ame extinguishes.
This method to lower h is not unique and the initial mixtures could be chosen otherwise. However, as argued in section 6.3.2, the exact choice of the starting points is not crucial.
The resulting data set is shown in gure 6.5 for a stoichiometric methane/air mixture, where YO2
and h have been appointed as controlling variables. The mass fraction of oxygen is chosen as progress
variable, because it is continuously decreasing in these amelets. The progress variable is scaled using
the initial and equilibrium value YO2 , and YO2 ,eq of the 1D manifold (T = 300 K):
Y=
YO2 YO2 ,
.
YO2 ,eq YO2 ,
(6.28)

97
99
0.0
hi
0.4
0.6
i +1
0.8
1.0
equilibria
(Y , h )
n
rsio
nve
l co
fue
h (kJ g1 )
0.2
h i+1
1.2
0
0.2
0.4
0.6
0.8
Y
Figure 7.5: Projection the two-dimensional FGM data set onto the Yh plane for a stoichiometric
Figure 6.5: Projection ofof the two-dimensional FGM data set onto the Yh plane for a stoichiometric
methane/air mixture. The bold curve indicates the 1D manifold. the circle a a detail the underlying amelets
methane/air mixture. The bold curve indicates the 1D manifold. InIn the circle detail ofof the underlying amelets
is is magnied.
magnied.
the initial and equilibrium value YO2 , and YO ,eq of the 1D manifold (T = 300 K):
Since at low enthalpy the mass fraction of O2 can2 be smaller than YO2 ,eq at h = hu , the value of Y can
YO2 is conserved in the amelet equations, the initial
be larger than 1 (see gure 6.5). Because enthalpy YO2 ,
Y=
.
(7.28)
and equilibrium value of h are the same in eachYamelet. O2 ,
O2 ,eq Y However, due to dierential diusion eects
the enthalpy changes locally in the amelets. This be smaller thanthe element= h , the value of Y can
Since at low enthalpy the mass fraction of O2 can also holds for YO2 ,eq at h mass fractions, which
u
are governed by1a(see gure 7.5). Because enthalpy Since these local variations are included ininitial
be larger than similar equation as the enthalpy. is conserved in the amelet equations, the the
manifold, additional controlling the same in each amelet.to model dierential diusiondiffusionwhile
and equilibrium value of h are variables are not needed However, due to differential eects, effects
inthe enthalpy changes locally inbe accounted for by using additional controllingmass fractions, which
ILDM these changes can only the amelets. This also holds for the element variables.
In literature several techniques to as the enthalpy. Since these proposed, ranging included in the
are governed by a similar equation store manifold data have beenlocal variations are from articial
neural networks [4] and in situ adaptive tabulation [23] to orthogonal polynomials [30]. The most
manifold, additional controlling variables are not needed to model differential diffusion effects, while
important criteria for judgingonly bestorage techniquesusingthe memory required for the storage of
in ILDM these changes can these accounted for by are additional controlling variables.
the table, the CPU time required to retrieve a value, and the accuracy of the retrieved value. articial
In literature several techniques to store manifold data have been proposed, ranging from In the
following a method is and in situwhich exploits the characteristics of the polynomials [30]. in a fast,
neural networks [4] explained adaptive tabulation [23] to orthogonal data set, resulting The most
accurate, and stable look-up procedure.
important criteria for judging these storage techniques are the memory required for the storage of
the table, the CPU time required to retrieve a value, , that are needed of the retrieved simulations,
For each separate amelet all dependent variables and the accuracy in combustion value. In the
following a method is explained which exploits the characteristics To realize this, the Y in a is
and the enthalpy are stored as a function of the progress variable. of the data set, resultingrange fast,
accurate, and stable look-up procedure.
divided into equal-sized regions, and a cubic polynomial is tted to the data points in each region
For each separate amelet consisting of the separate cubic polynomials, and its rst derivative
(see gure 6.6). The resulting t,all dependent variables , that are needed in combustion simulations,
and the to be continuous at the function of the progress variable. To the inlet and equilibrium
are forcedenthalpy are stored as a region boundaries. In order to predict realize this, the Y range is
divided into equal-sized t matches a data polynomial both compositions. The accuracy region
composition accurately, theregions, andthecubic set exactly atis tted to the data points in eachof the
t(see gure 7.6). The resulting t, consisting of regions or the order of the polynomials. rst derivative
can be improved by increasing the number of the separate cubic polynomials, and its
are forced to be continuous at the region boundaries. In order to predict the inlet and equilibrium
To retrieve data from the manifold for given values of the controlling variables the following
composition accurately, the t matches the data set exactly at both compositions. The accuracy of the
procedure is followed. First it is determined where the entry (Y , h ) is located. Therefore, the
t can be improved by increasing the number of regions or the order of the polynomials.
enthalpy is evaluated in two succeeding amelets at Y = Y , yielding hi and hi+1 for amelets i and
To retrieve data from the manifold for given values of the controlling variables the following
i + 1 such that hi+1 < h < hi (cf. gure 6.5). In the next step, intermediate values for the dependent
procedure is followed. First it is determined where the entry (Y , h ) is located. Therefore, the
variables are evaluated in both amelets: i and i+1 . The nal value is obtained using a linear
Chapter 7 J. A. van Oijen

98
100
3 a (1) Yi
i=0 i
3 a (2) Yi
i=0 i
1/3
2/3
3 a (3) Yi
i=0 i
Y
Figure 7.6: Piece-wise polynomial t (line) to the data points (dots). In this example the Y range is divided in
Figure 6.6: Piece-wise polynomial t (line) to the data points (dots). In this example the Y range is divided in
regions and in each region the data is tted with cubic polynomial.
33regions and in each region the data is tted with aacubic polynomial.
enthalpy is between in two succeeding amelets at Y = Y , yielding h i and h i+1 for amelets i and
interpolationevaluated these values:
i + 1 such that h i+1 < h < h i (cf. gure 7.5). In the next step, intermediate values for the dependent
h
h and
h The
variables are evaluated in both amelets: ii+1 i i+1 .i h nal value is obtained using a linear
=
+ i
i+1 .
(6.29)
interpolation between these values: hi hi+1
h hi+1
Due to this linear interpolation as a function of h, the rst derivative of with respect to h is discontin
i+1
h i h i+1
a h h
uous at the amelets, which maybe= problem in i +
gradient-based solvers. This problem can be solved
.
(7.29)
i+1
h i h i+1
h i dependent variables change very smoothly
by using a higher-order interpolation. However, since the h
in the direction of h, this imperfection has no consequences in our case.
Due to this linear interpolation as a function of h, the rst derivative of with respect to h is disOne of the main advantages of this storage technique is that the boundaries of the tabulated domain
continuous at the amelets, which may be a problem in gradient-based solvers. This problem can
are known very accurately. Therefore, eective measures can be taken if an entry is outside this
be solved by using a higher-order interpolation. However, since the dependent variables change very
domain, resulting in a robust look-up method. Such a measure is needed, for instance, when burnt
smoothly in the direction of h, this imperfection has no consequences in our case.
gases are cooled intensively, e.g. due to a local heat sink. The enthalpy of the mixture then drops
One of the main advantages of this storage technique is that the boundaries of the tabulated domain
below that of the coldest amelet, and the composition enters the so-called low-enthalpy region,
are known very accurately. Therefore, effective measures can be taken if an entry is outside this
shown in gure 6.7. Since no amelet data is available there, the manifold must be continued into
domain, resulting in a robust look-up method. Such a measure is needed, for instance, when burnt
this region. A similar problem occurs in ILDM: at low temperatures the steady-state equations do
gases are cooled intensively, e.g. due to a local heat sink. The enthalpy of the mixture then drops
not yield a solution and the manifold is continued into the cold region by linear extrapolation [14].
below that of the coldest amelet, and the composition enters the so-called low-enthalpy region,
Recently, Gicquel et al. [7] proposed a method very similar to FGM called ame prolongation of
shown in gure 7.7. Since no amelet data is available there, the manifold must be continued into
ILDM (FPI) to tackle this problem. They prolongated the ILDM into the cold region by using the
this region. A similar problem occurs in ILDM: at low temperatures the steady-state equations do
compositions in a 1D ame. Eectively, the 1D ILDM becomes exactly the same as the 1D FGM
not yield a solution and the manifold is continued into the cold region by linear extrapolation [14].
by using this prolongation technique. Application of FPI to multi-dimensional manifolds with two or
Recently, Gicquel et al. [7] proposed a method very similar to FGM called ame prolongation of
more progress variables is not discussed in [7].
ILDM (FPI) to tackle this problem. They prolongated the ILDM into the cold region by using the
In FGM the a 1D ame. Effectively, the the low-enthalpy region as follows. First 1D FGM
compositions in manifold is prolongated into 1D ILDM becomes exactly the same as the the lowenthalpy region is mapped onto the unit square (, ) = [0,to multi-dimensional manifolds with two or
by using this prolongation technique. Application of FPI 1] [0, 1] by using
more progress variables is not discussed in [7].
YY
hh
In FGM the manifold is =
prolongated the low-enthalpymin ,
into and =
region as follows. First the low(6.30)
Yeq Y
hL hmin

99
101
0.0
0.5
coldest amelet
=1
1.5
=1
equilibria
h (kJ g1 )
1.0
=
0
ini
tia
l
2.0
2.5
3.0
=0 A
0
0.2
0.4
0.6
0.8
Y
Figure 7.7: Interpolation the low-enthalpy region.
Figure 6.7: Interpolation inin the low-enthalpy region.
enthalpy region enthalpy onto coldest amelet and = 0, 1 0, 1 by of the

where hL () is theis mapped at the the unit square (, ) hmin the enthalpy using chemical equilibrium
with T = T min (point A in gure 6.7). The proles of the dependent variables along the coldest
Y Y
h h min
amelet (, 1), the initial mixtures (0, ) and the and =equilibria (1, ) are known. The values
chemical
=
,
(7.30)
Yeq Y by
h L h min
in the low-enthalpy region are approximated interpolation using these proles:
where h L () is the enthalpy at the coldest amelet and h min () + (1 )g()the chemical equilibrium
(, ) = (, 0) + (, 1) (, 0) f the enthalpy of ,
(6.31)
with T = Tmin (point A in gure 7.7). The proles of the dependent variables along the coldest
amelet (, 1), the initial mixtures (0, ) and the chemical equilibria (1, ) are known. The
with
(1, ) approximated
(0, ) using these
values in the low-enthalpy region are (1, 0) and by interpolation (0, 0) . proles:
f () =
g() =
(6.32)
(1, 1) (1, 0)
(0, 1) (0, 0)
(, = (, 0) + 0), linear (, 0)] are used (1 )g( )] ,
When (1, 1) = (1, 0) or)(0, 1) = (0,[(, 1) relations [ f ( ) + instead of (6.32), i.e. f () (7.31)
=
and g() = . Since in most practical cases extensive cooling takes place in the burnt gases after the
with
ame front, most entries in the low-enthalpy region are very close to) chemical equilibria ( = 1),
(1, ) (1, 0)
(0, (0, 0)
and g( ) =
.
(7.32)
f ( ) =
where almost no interpolation error is made. Although this procedure to continue the manifold in the
(1, 1) (1, 0)
(0, 1) (0, 0)
low-enthalpy region is rather ad-hoc and might need some more attention, it appears to work quite
When
well. (1, 1) = (1, 0) or (0, 1) = (0, 0), linear relations are used instead of (7.32), i.e. f ( ) =
and g( ) = . Since in most practical cases extensive cooling takes place in the burnt gases after the
ame front, most of equations
6.4.2 Reduced setentries in the low-enthalpy region are very close to chemical equilibria ( = 1),
where almost no interpolation error is made. Although this procedure to continue the manifold in the
After the FGM region is rather ad-hocitand might needto a standardattention, it appears to work quite
low-enthalpy is stored in a database, can be linked some more CFD code. This means that together with the momentum and continuity equations, the CFD code has to solve dierential equations
well.
for the controlling variables. In ILDM these conservation equations for the controlling variables are
derivedReduced set of of the full system onto the manifold. The dierent terms in the governing equa4.2 by a projection equations
L
tions are projected using the left eigenvectors sfast corresponding to the fast chemical processes [14].
After the FGM is stored in a database, it can be and the a standard CFD code. This means that toDue to the denition of the ILDM, the convection linked tochemical source terms remain unchanged
gether the the momentum aected by the projection. The projection of the diusion term equations
and onlywithdiusion term is and continuity equations, the CFD code has to solve differentialin ILDM
isfor the controlling variables. In ILDMwhere conservation equations for the controllingrestricted to
schematically shown in gure 6.8 (a), these it can be seen that the diusion term D is variables are
102
Rst
R
L
sfast
C
P
+
Rst
D
+
(a) Rst + Rst = 0
(b) C + D + R = 0
Figure 6.8: Projection of the dierent terms onto the manifold (dashed line) in the ILDM (a) and FGM (b)
method. The dierent terms and their projections are represented by black and gray arrows, respectively.
+
the manifold due to the balance between the large chemical production Rst and consumption Rst terms
in the steady-state relation. Since the burning velocity of premixed laminar ames is determined by a
balance between reaction and diusion, an accurate projection of the diusion terms is important. Although the eigenvector projection is mathematically the only correct one, other (simplied) projection
techniques have been used successfully [5, 6, 8].
As shown in gure 6.8 (b), the steady-state relations in the FGM method are given by a balance
between 1D convection C, diusion D and reaction R: C + D + R = 0. The vector P represents
the multi-dimensional perturbations in (6.17), which are projected onto the manifold by the large
terms C, D and R. The main dierence with ILDM is that in FGM the steady-state relations (6.24)
hold for all species, eectively freezing the mixture composition. Therefore, the multi-dimensional
perturbations are not only projected onto the manifold, but in the manifold as well. This results in
P = 0, which is of course related to the assumption that the multi-dimensional perturbations are
neglected. If a ame is extremely stretched or curved locally, the perturbations cannot be neglected
and the dimension of the manifold should be increased, reducing the component of P perpendicular
to the manifold.
Since in the FGM method all species are assumed to be in a 1D steady state, a projection of
the full set of equations onto the manifold does not result in dierential equations for the controlling
variables like in ILDM. Therefore, ordinary conservation equations for the controlling variables Y
and h are used in FGM. The dierential equation for Y reads
(Y)
+ (uY)
Y = Y ,
t
LeY c p
(6.33)
which does include the multi-dimensional perturbation for the progress variable PY . Because the
multi-dimensional perturbations depend on the choice of progress variable, the nal result also depends on this choice. This is not the case in ILDM if the eigenvector projection is applied. However,
if the multi-dimensional perturbations are small, the dierences between results computed with different progress variables will be negligible.
The conservation equation for the enthalpy is given by
(h)
+ (uh)
h = H,
t
cp
(6.34)
103
where H represents the enthalpy ux due to dierential diusion:
H=
cp
Ns
1
1 hi Yi ,
Lei
i=1
(6.35)
being a function of the species mass fraction gradients. Application of the chain rule yields
H=
cp
Ns
i=1
Yi
1
Yi
1 hi
Y+
h ,
Lei
Y
h
(6.36)
stating that H depends on the gradients of Y and h. This may lead to unrealistic energy uxes through
inert walls, where Yi = 0 but h 0. This problem is caused by the introduction of a manifold and
n
n
is thus present in all manifold techniques. In FGM this problem can be circumvented by assuming
that locally the enthalpy is given by the amelet relation h = h1D (Y), which is a function of Y only.
This approximation is related to the assumption that locally no heat losses occur. Using this relation
between h and Y, the enthalpy ux H can be written as a function of the gradient of Y only:
H = c Y,
with
c=
cp
Ns
i=1
Yi Yi h1D
1
1 hi
+
.
Lei
Y h Y
(6.37)
The coecient c is stored in the FGM database together with the variables needed to solve the conservation equations, i.e. , , c p , T and Y . This is obviously much more ecient and accurate than
storing the species mass fractions and computing these variables at run-time.
6.5
Test results
To demonstrate the performance of the FGM method, two test cases of premixed laminar methane/air
ames have been simulated by using a 2D manifold as presented in the previous section. Results are
shown for one and two-dimensional burner-stabilized ames, and they are compared with results obtained with the reaction mechanism from Smooke [27], which consists of 16 species and 25 reversible
reactions.
6.5.1
One-dimensional validation
Apart from numerical inaccuracies, the most elementary premixed ame the freely propagating
at ame should be reproduced exactly by FGM. Burner-stabilized 1D ames, however, are subjected to non-adiabatic eects due to heat losses to the burner and are therefore more challenging.
The burner outow is positioned at x = 0 and the burner area x < 0 is kept at a constant temperature of T burner = 300 K. Stationary solutions are computed for dierent values of the mass burning
rate m ranging from 0.05 to 0.40 kg m2 s1 . The adiabatic mass burning rate of this stoichiometric methane/air ame is m = 0.421 kg m2 s1 , which corresponds to a laminar burning velocity of
sL = 0.375 m s1 . The ames are computed with the 1D ame solver CHEM1D [3] developed at
the Eindhoven University of Technology. This code uses an exponential nite-volume discretization
in space and the resulting system is solved using a fully implicit, modied Newton technique [28].
Adaptive gridding is implemented to increase the resolution around the ame front.
For m = 0.30 kg m2 s1 the proles of the controlling variables YO2 and h are shown in gure 6.9.
It can be seen that the results of the FGM method are in excellent agreement with those of the detailed
Chapter 7 J. A. van Oijen

102
104
0.30
1.5
0.35
1.0
0.40
h (J kg1 )
h (J kg1 )
0.25
2.0
YO22 10
YO 10
2.5
0.45
0.5
2
O2
0.0
1
2
x (mm)
0.50
O
Figure 7.9: Proles of the controlling variables O2 2 and in one-dimensional burner-stabilized ame with
Figure 6.9: Proles of the controlling variables YYO2and hhin a a one-dimensional burner-stabilized ame with
2
m = 0.30 kg2 2 s1 . Line: detailed chemistry; symbols: FGM. symbols do not represent the computational
ms1 . 1
m = 0.30 kg m
Line: detailed chemistry; symbols: FGM. The The symbols do not represent the computational grid; ner grid has been used in the calculations.
grid; a much a much ner grid has been used in the calculations.
2.5
Yii 103
Y 103
2.0
1.5
1.0
H 10
0.5
2
CH2 O
0.0
1
2
x (mm)
Figure 7.10: Proles of the radicals YCH2 O and YH in a one-dimensional burner-stabilized ame with m =
CH2
Figure 6.10: Proles of the radicals YCH2 OOand YHHin a one-dimensional burner-stabilized ame with m =
2 s1
0.30 kg m211 . Line: detailed chemistry; symbols: FGM. The symbols do not represent the computational
0.30 kg m2 s . Line: detailed chemistry; symbols: FGM. The symbols do not represent the computational
grid; much ner grid is used in the calculations.
grid; a a much ner grid is used in the calculations.

103
105
1.3
5.6
1.2
5.4
1.1
1.0
5.2
0.9
5.0
0.8
4.8
d (mm)
1/Tb (104 K1 )
5.8
0.7
4.6
0.6
4.4
0.5
4.2
0.4
0.05
0.10
0.20
0.40
m (kg m2 s1 )
Figure 7.11: Arrhenius plot for the ame temperature T and the stand-o distance d as function of the mass
Figure 6.11: Arrhenius plot for the ame temperature T b b and the stand-off distance d as function of the mass
burning rate (log scale). Line: detailed chemistry; symbols: FGM.
burning rate mm (log scale). Line: detailed chemistry; symbols: FGM.
in space and the resulting system is solved conserved in implicit, modied Newton technique the
reaction mechanism. Although the enthalpy is using a fully the amelets that are used to construct [28].
Adaptive gridding is drop due to to increase the resolution modelled ame front.
manifold, the enthalpyimplementedheat losses in the burner isaround the correctly with FGM. Beside
2 1
the burner, the large gradients in h in O2 ame front due to Lewisthese For m = 0.30 kg m sin the proles of the controlling variables Ytheand h are shown in gure 7.9.
non-adiabatic eects
It can eects (0 < x results of the reproduced very in excellent agreement with CH of the H are
numberbe seen that the< 1 mm) are FGM method are well. The mass fractions ofthose2 O anddetailed
reaction as function Although the enthalpy for these radicals, which are very dicult to reproduce
displayed mechanism.of x in gure 6.10. Evenis conserved in the amelets that are used to construct the
manifold, the enthalpy drop due to heat losses in the burner is modelled correctly with FGM. Beside
accurately using ILDM, the resemblance between the FGM and detailed results is striking.
these non-adiabatic effects in the burner, the large gradients in h in the ame front due to LewisIn gure 6.11 two global observables of 1D burner-stabilized ames are presented as function of
number effects (0 < x < 1 mm) are reproduced very well. The mass fractions of CH2 O and H are
the mass burning rate. These global observables are the non-adiabatic ame temperature T b and the
displayed as function of x in gure 7.10. Even for these radicals, which are very difcult to reproduce
stand-o distance d, here dened as the position where the absolute value of the chemical source term
accurately using ILDM, the resemblance between the FGM and detailed results is striking.
|O2 | reaches its maximum. The ame temperature is measured in the burnt mixture at x = 10 cm, and
In gure 7.11 two global observables of 1D burner-stabilized ames are presented as function of
is presented in the form of an Arrhenius plot. The dierences in the ame temperature are smaller than
the mass burning rate. These global observables are the non-adiabatic ame temperature Tb and the
5 K and, therefore, almost invisible in the gure. Although the results for the stand-o distance are
stand-off distance d, here dened as the position where the absolute value of the chemical source term
somewhat poorer, the deviation is within 1 percent of the ame thickness and therefore, also hardly
|O2 | reaches its maximum. The ame temperature is measured in the burnt mixture at x = 10 cm, and
visible in gure 6.11. Because at low mass burning rates the deviation from an adiabatic amelet
is presented in the form of an Arrhenius plot. The differences in the ame temperature are smaller than
increases, the dierence between the FGM and detailed computations increases as well.
5 K and, therefore, almost invisible in the gure. Although the results for the stand-off distance are
somewhat poorer, the deviation is within 1 percent of the ame thickness and therefore, also hardly
visible in gure 7.11. Because at
6.5.2 Two-dimensional validationlow mass burning rates the deviation from an adiabatic amelet
increases, the difference between the FGM and detailed computations increases as well.
The potential of the FGM method is demonstrated by a calculation of a 2D burner-stabilized laminar
premixed ame. In this test case almost all the features found in premixed laminar ames such
5.2 Two-dimensional validation
as ame cooling, dierential diusion, stretch and curvature are present. We have simulated a
The potential of with an method is demonstrated by a which stabilizes on a 2D slot burner in a
methane/air amethe FGM equivalence ratio of = 0.9calculation of a 2D burner-stabilized laminar
premixed ame. In this test case almost all the features found shown in gure 6.12. The such
box. The computational domain of the burner conguration is in premixed laminar ames burner
as is 6 mm wide, while the box is 24 mm wide. The burner box walls We have simulated
slot ame cooling, differential diffusion, stretch and curvature and are present.are kept at a constant a
methane/air ame = 300 equivalence ratio of the inlet is parabolic with a 2D slot burner in
temperature of T wall with anK. The velocity prole at= 0.9 which stabilizes on a maximum velocity a
box. = 1.0 m s1 .
of vmax The computational domain of the burner conguration is shown in gure 7.12. The burner
106
Constanttemperature wall
21 mm
Symmetry plane
Outlet
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
12 mm
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
3 mm
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
11111111111
00000000000
Inlet
Figure 6.12: Numerical conguration used for the 2D computation. Results are shown for the region enclosed
by the dotted line.
Isocontours of T and the mass fractions of O2 , CO, H, CH2 O and O are shown in gure 6.13 on a
portion of the computational domain for both the detailed and reduced chemistry computations. It can
be seen that the results obtained with the FGM method are in excellent agreement with the detailed
chemistry computations: not only the position of the ame front is predicted very well, but the absolute
values of the mass fractions are reproduced as well. Flame cooling governing the stabilization of the
ame on the burner is captured very well by FGM, although one can hardly speak of amelets in
this cold region. Also in the ame tip, where stretch and curvature are very important, the reduced
chemistry computations appear to coincide with the detailed chemistry calculations. Close to the
burner wall there are some dierences in the YCO contours: the contours do not attach to the wall
in the FGM case. Since YCO follows directly from the manifold database, which does not include
diusion along the ame front, the CO concentration is under-predicted by FGM in this diusiondominated area. To take this into account the manifold should be extended with a second progress
variable.
The main dierence between the detailed and reduced chemistry computations is the computation
time. The computation with detailed chemistry took approximately two weeks to converge, while the
FGM results were obtained within a few hours. These gures are only a rough indication of the gain
in computation time, because dierent numerical solvers, grids and initial elds have been used. In
the next section a more accurate study of the numerical eciency of FGM is discussed.
Detailed chemistry
107
FGM
Detailed chemistry
FGM
0.5
56(33)188 103
O2
400(200)2000
0.5
-0.5
0.5
-0.5
0.5
1
7(7.5)37 103
CO
0.5
3.2(3.2)16 105
0.5
-0.5
0.5
-0.5
0.5
CH2 O
4.5(4.125)21 105
0.5
37(38)189 105
0.5
-0.5
0.5
-0.5
0.5
Figure 6.13: Isocontours of T , YO2 , YCO , YH , YCH2 O and YO computed using detailed chemistry (left) and FGM
(right). The contours are drawn at equal intervals covering the entire range of each variable (indicated in the
upper right corner of each graph). The same isolevels are used for detailed and FGM computations. The space
coordinates are given in cm.
108

Mixture
H2 /air
CH4 /air
Chemistry model
Detailed
Ns = 7, Nr = 7
ILDM
Npv = 1, Nav = 1
FGM
Npv = 1, Nav = 1
Detailed
Ns = 16, Nr = 25
ILDM
Npv = 1, Nav = 1
FGM
Npv = 1, Nav = 1
FGM explicit Npv = 1, Nav = 1
tstep (ms)
226
110
112
657
110
9
ttotal (s)
481
63
62
998
46
21
Error (%)
0
8.63
0.08
0
0.05
0.05
Table 6.1: Results of time-dependent 1D ame simulations using dierent reaction models. The computation
time per time step, tstep , the total simulation time, ttotal , and the error (6.38) are shown.
6.5.3
Computational eciency
Besides the accuracy, the eciency is another important aspect of reduction methods. In order to give
a further indication of the eciency of FGM, the computation times of detailed and reduced chemistry
simulations with ILDM and FGM are compared. For all models we determined the time needed to
perform a time-dependent ame simulation at the same conditions. 1D burner-stabilized ames have
been modelled for a period of 103 s. At t = 0 the mass ow rate has been reduced from 90% to
85% of the adiabatic mass burning rate. To solve the equations we have used the fully implicit solver
CHEM1D [3] with adaptive time stepping. The FGM method has also been used with an explicit time
stepper using constant time steps.
Two dierent mixtures are considered: a stoichiometric hydrogen/air and methane/air mixture.
The hydrogen reaction mechanism contains 7 species and is a subset of the methane reaction mechanism from [27]. The manifolds used are 2D with one progress variable and the enthalpy as additional
controlling variable. The same look-up procedure is used for both manifold methods. Lewis numbers
equal to 1 are used in order to prevent variations in the element mass fractions. The computations are
performed on a Silicon Graphics R10000 workstation and the results are shown in table 6.5.3.
For the hydrogen/air mixture the CPU time per time step tstep reduces a factor 2 when a manifold is
applied. This speed up is caused by the reduction of the number of dierential equations to be solved.
For the detailed chemistry simulation of the methane/air mixture, tstep is 3 times larger than for the
hydrogen/air mixture, because approximately 3 times as much dierential equations and reaction rates
have to be computed. For the reduced chemistry computations, however, tstep is independent of the
number of species and reactions in the detailed reaction model. Therefore, the speed up is almost a
factor 6 for the methane/air mixture. In this case the speed up is not only caused by a smaller number
of equations to be solved, but by a faster evaluation of the chemical source terms as well. For the
simple hydrogen mechanism with only 7 reactions, the direct computation of the chemical source
term is as fast as the retrieval from the manifold database.
Another advantage of using reduced models is that larger time steps can be taken, because the
smallest time scales have been eliminated. Therefore, the total CPU time ttotal of the hydrogen/air
computations using reduced chemistry is 8 times less than the detailed chemistry simulation. For the
methane/air ame, the total speed up becomes a factor 20. An even higher eciency is reached if
an explicit time-integration scheme is used for the reduced chemistry computations. In that case the
CPU time is reduced by almost a factor of 50 compared to the methane/air simulation using detailed
chemistry. For more complex reaction mechanisms (GRI-mech: Ns = 49, Nr = 279 [26]) and multidimensional systems the speed up will be even larger.
109
Since the same storage and retrieval method is used for the ILDM and FGM computations, the
CPU times are almost the same. The results computed with the 2D FGM are, however, more accurate
than the results computed with the 2D ILDM. To have an indication of the accuracy, the time evolution
of the enthalpy h(0, t) at the burner edge is compared with results obtained with detailed chemistry.
The error is dened as the absolute dierence with the enthalpy hdet computed with detailed chemistry
averaged over the total simulation period t, viz.
1
Error =
th
|h(0, t) hdet (0, t)| dt.
(6.38)
The error is scaled with the total variation in the enthalpy h = hdet (0, 0) hdet (0, t). For the
hydrogen/air mixture the error in the ILDM computation is almost 9%. This error is mainly caused
by the error in the adiabatic burning velocity, which is 8% too low in the ILDM case. The error of
the FGM computations is smaller than 0.1% for both gas mixtures. For the methane/air mixture it is
not possible to perform an ILDM computation, because an applicable 2D ILDM cannot be generated.
Only in a small region around chemical equilibrium, enough chemical processes can be assumed in
steady state. A 3D ILDM with two progress variables can be found for temperatures higher than
1200 K, but the accuracy is still too low and the mass burning rate is overestimated by a factor 2 [5].
Three progress variables and the enthalpy would result in a 4D manifold. Besides that these highdimensional manifolds require a large amount of memory and are less ecient than low-dimensional
manifolds.
The computation of the 2D FGM databases used in these simulations involved approximately 30
minutes, which is quite long compared to the CPU times listed in table 6.5.3. Obviously, it is not
ecient to construct a FGM for a single 1D ame simulation. However, the computation time that
can be gained by using FGM in a series of multi-dimensional ame simulations is orders of magnitude
larger than the time needed to construct the database.
Another important numerical aspect of ame simulations is the computational grid. Related to the
wide range of time scales, there is also a large dierence in length scales in premixed laminar ames.
In order to resolve the thin reaction layer, where peaks in the dierent radical concentrations occur
(cf. gure 6.10), many grid points with ne spacing are required to get an accurate representation of
the solution. When FGM is applied, these small scales do not have to be resolved, because the ame is
described by the relatively slowly varying progress variable. To demonstrate this, adiabatic 1D ames
have been computed on grids with a dierent number of grid points. In gure 6.14 the relative error
in the mass burning rate is plotted as function of the number of grid points Np in the ame front. The
relative error is dened as (mNp m)/m, where mNp is the mass burning rate computed with Np grid
points in the ame front. The value of m is found by Richardson extrapolation.
The detailed chemistry and FGM results show the same behaviour: for large Np the error decreases
quadratically with the number of grid points, indicating that the space discretization is second order.
At lower values of Np the behaviour is only rst order, which is typical for the discretization scheme
of Thiart [29] used in the code. Because the smallest scales do not have to be resolved, the error of
the FGM computations is approximately 50% smaller than detailed chemistry computations with the
same number of grid points. In other words, when FGM is applied, 25 percent less grid points are
required to reach the same accuracy as in the detailed chemistry computations.
Chapter 7 A. van Oijen

Chapter 6 J. J.A. van Oijen
108
110
101
|m Np m|/m
102
103
104
105
10
100
Np
1000
Figure 7.14: Relative error in in the mass burning rate (m m)/m as function of the number of grid points p
Figure 6.14: Relative error the mass burning rate (m Np Np m)/m as function of the number of grid points N Np
inin the ame front. Open symbols: detailed chemistry; lled symbols: FGM. The line indicates slope ofof 2.
the ame front. Open symbols: detailed chemistry; lled symbols: FGM. The line indicates a a slope 2.
6 6.6 Conclusions
Conclusions
InIn this chapter the FGM method has been introduced and it has been used to compute burner-stabilized
this chapter the FGM method has been introduced and it has been used to compute burner-stabilized
premixed laminar ames. Since manifold is generated by using amelets in FGM, it can be conpremixed laminar ames. Since a a manifold is generated by using amelets in FGM, it can be considered a combination of a manifold and a amelet approach. It combines the advantages of both
sidered asas a combination of a manifold and a amelet approach. It combines the advantages of both
approaches. FGM shares the basic idea with amelet methods that multi-dimensional ame can be
approaches. FGM shares the basic idea with amelet methods that a a multi-dimensional ame can be
regarded a a set 1D ames. It is assumed that the structure of the ame front is similar in both types
regarded asas set ofof 1D ames. It is assumed that the structure of the ame front is similar in both types
ofof ames and that the mixture composition is governed by balance between 1D convection, diffusion
ames and that the mixture composition is governed by a a balance between 1D convection, diusion
and reaction. Since the main parts convection and diffusion are included in the amelet equations,
and reaction. Since the main parts ofof convection and diusion are included in the amelet equations,
the FGM more accurate in regions where reaction is balanced by convection and diffusion than
the FGM is is more accurate in regions where reaction is balanced by convection and diusion than
methods based on local chemical equilibria. high-temperature regions, where chemistry is domimethods based on local chemical equilibria. InIn high-temperature regions, where chemistry is dominant, the amelets are attracted by the ILDM and the ILDM and FGM are identical. Because FGM
nant, the amelets are attracted by the ILDM and the ILDM and FGM are identical. Because FGM
is is more accurate in low-temperature regions, less controlling variables are needed, which makes
more accurate in low-temperature regions, less controlling variables are needed, which makes
it an ecient reduction method. Test results of burner-stabilized methane/air ames show that only
it an efcient reduction method. Test results of burner-stabilized methane/air ames show that only
one progress variable apart from the enthalpy sufcient to reproduce detailed chemistry simulations
one progress variable apart from the enthalpy is is sucient to reproduce detailed chemistry simulations
very well. comparison with results obtained using a similar 2D ILDM is not possible, because for
very well. AA comparison with results obtained using a similar 2D ILDM is not possible, because for
methane/air mixtures an ILDM with one progress variable can only be found a a small region around
methane/air mixtures an ILDM with one progress variable can only be found inin small region around
chemical equilibrium. This makes the application this ILDM in ame simulations practically imchemical equilibrium. This makes the application ofof this ILDM in ame simulations practically impossible. When two progress variables are used, usable ILDM can be constructed. However,
possible. When two progress variables are used, a a usable ILDM can be constructed.However, it it
overestimates the mass burning rate by a factor two [5], which would probably result in ash-back
overestimates the mass burning rate by a factor ofof two [5], which would probably result in ash-back
for the 2D ame considered this chapter, while the FGM with one progress variable predicts the
for the 2D ame considered inin this chapter, while the FGM with one progress variable predicts the
correct position the ame front.
correct position ofof the ame front.
The implementation ofof FGM in a ame simulation is typical for a manifold method.The reacThe implementation FGM in a ame simulation is typical for a manifold method. The reaction rates and other essential variables are stored inin database, which is is used to solve the dierential
tion rates and other essential variables are stored a a database, which used to solve the differential
equations for the controlling variables. Because ordinary conservation equations are solved, the imequations for the controlling variables. Because ordinary conservation equations are solved, the implementation inin CFD codes is relatively easy compared with amelet methods. In existing amelet
plementation CFD codes is relatively easy compared with amelet methods. In existing amelet
111
methods a kinematic equation for the scalar G is solved [20, 33]. In this so-called G-equation the
burning velocity enters explicitly and the inuence of ame stretch and curvature have to be modelled. Moreover, while the conservation equations for the controlling variables are valid throughout
the complete domain, the G-equation is only valid at one value G = G0 , which denotes the position
of the ame sheet. Everywhere else G is simply dened as the distance to the ame sheet (| G| = 1),
resulting in a ame of constant thickness. In FGM the ame thickness is in general not constant and,
as the burning velocity, it follows from the conservation equations for the controlling variables. A disadvantage of solving conservation equations for the controlling variables, however, is that the ame
front needs to be resolved. This requires a large number of grid points compared to amelet methods
which do not have to resolve the internal structure of the ame front.
An advantage of FGM is that the number of progress variables is not limited to one as in existing
amelet approaches. Although the test results show that one progress variable and the enthalpy are
sucient to reproduce detailed chemistry simulations very well, more progress variables can be added
to increase the accuracy of the method. Addition of progress variables might cause problems in the
look-up procedure as described in section 6.4.1, because the dierent amelets converge to form a
lower-dimensional manifold (cf. gure 6.4). However, this problem can be solved by using a modied
look-up technique. When the dierence between the amelets is too small (and the dimension of the
manifold has actually decreased locally), the values are retrieved from the amelet(s) in the lowerdimensional manifold.
A disadvantage of FGM compared to ILDM is that a general procedure to create multi-dimensional
manifolds cannot be derived. Since the researcher has to determine the starting points of the amelets,
the construction of a FGM cannot be automated to the same degree as that of an ILDM, where only
the dimension of the manifold has to be prescribed by the researcher.
The test results presented in this lecture demonstrate the enormous potential of FGM. Results
of computations with detailed chemistry have been reproduced very accurately within a fraction of
the computation time. Since a FGM with one progress variable reaches the same accuracy for a
methane/air ame as an ILDM with at least 3 dimensions, the dimension of the look-up database can
be low. This not only reduces the computation time, but it also makes the numerical implementation
relatively easy. The implementation of FGM in CFX-4 made it possible to simulate the combustion
process in a realistic furnace [16] including radiation.
A detailed investigation of the eects of the multi-dimensional perturbations on the accuracy of
FGM is given by Van Oijen [15]. The inuence of ame stretch and the deviations from purely
premixed behaviour are studied systematically in [15]. More recently, the FGM method has been
applied in DNS, LES, and RANS simulations of premixed and non-premixed turbulent ames [1, 2,
17, 18, 31].
References
[1] Albrecht B.A., Zahirovic S., Bastiaans R.J.M., van Oijen J.A. and de Goey L.P.H., A Premixed
Flamelet-PDF Model for Biomass Combustion in a Grate Furnace, Energy and Fuels, Accepted,
(2008)
[2] Bastiaans R.J.M., van Oijen J.A. and de Goey L.P.H., Application of Flamelet Generated Manifolds and Flamelet Analysis of Turbulent Combustion, Int. Jnl. Multiscale Comp. Eng. 4(3):307317, (2006)
112
[3] CHEM1D, A one-dimensional laminar ame code, Eindhoven University of Technology

http://www.combustion.tue.nl/chem1d
[4] Christo F.C., Masri A.R. and Nebot E.M., Articial neural network implementation of chemistry
with pdf simulation of H2 /CO2 ames, Combust. Flame 106:406427 (1996)
[5] Eggels R.L.G.M., Modelling of Combustion Processes and NO Formation with Reduced Reaction Mechanisms, Ph.D. thesis, Eindhoven University of Technology (1996)
[6] Eggels R.L.G.M. and de Goey L.P.H., Mathematically reduced reaction mechanisms applied to
adiabatic at hydrogen/air ames, Combust. Flame 100:559570 (1995)
[7] Gicquel O., Darabiha N. and Th venin D., Laminar premixed hydrogen/air counterow ame
e
simulations using ame prolongation of ILDM with dierential diusion, Proc. Combust. Inst.
28:19011908 (2000)
[8] Gicquel O., Th venin D., Hilka M. and Darabiha N., Direct numerical simulations of turbulent
e
premixed ames using intrinsic low-dimensional manifolds, Combust. Theory Modelling 3:479
502 (1999)
[9] de Goey L.P.H., Mallens R.M.M. and ten Thije Boonkkamp J.H.M., An evaluation of dierent
contributions to ame stretch for stationary premixed ames, Combust. Flame 110:5466 (1997)
[10] de Goey L.P.H. and ten Thije Boonkkamp J.H.M., A mass-based denition of ame stretch with
nite thickness, Combust. Sci. Technol. 122:399405 (1997)
[11] de Goey L.P.H. and ten Thije Boonkkamp J.H.M., A amelet description of premixed laminar
ames and the relation with ame stretch, Combust. Flame 119:253271 (1999)
[12] Lam S.H. and Goussis D.A., Conventional asymptotics and computational singular perturbation
for simplied kinetics modeling, in Reduced kinetic mechanisms and asymptotic approximations
for methane-air ames, edited by M.D. Smooke, pp. 227242, Springer Verlag, Berlin (1991)
[13] Maas U. and Pope S.B., Simplifying chemical kinetics: Intrinsic low-dimensional manifolds in
composition space, Combust. Flame 88:239264 (1992)
[14] Maas U. and Pope S.B., Laminar ame calculations using simplied chemical kinetics based on
intrinsic low-dimensional manifolds, Proc. Combust. Inst. 25:13491356 (1994)
[15] van Oijen J.A., Flamelet-Generated Manifolds: Development and Application to Premixed Laminar Flames, Ph.D. thesis, Eindhoven University of Technology (2002)
[16] van Oijen J.A., Lammers F.A. and de Goey L.P.H., Modelling of complex premixed burner
systems using amelet-generated manifolds, Combust. Flame 127:21242134 (2001)
[17] van Oijen J.A., Bastiaans R.J.M, Groot G.R.A and de Goey L.P.H., Direct numerical simulations
of premixed turbulent ames with reduced chemistry: Validation and amelet analysis, Flow
Turb. Combust. 75:67-84, (2005)
[18] van Oijen J.A., Bastiaans R.J.M and de Goey L.P.H., Low-Dimensional Manifolds in Direct Numerical Simulations of Premixed Turbulent Flames, Proc. Combust. Inst. 31:1377-1384, (2007)
113
[19] Peters N., Laminar amelet concepts in turbulent combustion, Proc. Combust. Inst. 21:1231
1250 (1986)
[20] Peters N., Turbulent Combustion, Cambridge University Press, Cambridge (2000)
[21] Peters N. and Rogg B. (editors), Reduced kinetic mechanisms for applications in combustion
systems, Springer Verlag, Berlin (1993)
[22] Peters N. and Williams F.A., Asymptotic structure of stoichiometric methane-air ames, Combust. Flame 68:185207 (1987)
[23] Pope S.B., Computationally ecient implementation of combustion chemistry using in situ
adaptive tabulation, Combust. Theory Modelling 1:4163 (1997)
[24] Pope S.B. and Maas U., Simplifying Chemical Kinetics: Trajectory-Generated Low-Dimensional
Manifolds, Technical Report FDA 93-11, Cornell University (1993)
[25] Schmidt D., Blasenbrey T. and Maas U., Intrinsic low-dimensional manifolds of strained and
unstrained ames, Combust. Theory Modelling 2:135152 (1999)
[26] Smith G.P., Golden D.M., Frenklach M., Moriarty N.W., Eiteneer B., Goldenberg M., Bowman
C.T., Hanson R.K., Song S., Gardiner Jr. W.C., Lissianski V.V. and Qin Z.
http://www.me.berkeley.edu/gri mech/
[27] Smooke M.D. and Giovangigli V., Formulation of the premixed and nonpremixed test problems,
in Reduced kinetic mechanisms and asymptotic approximations for methane-air ames, edited
by M.D. Smooke, pp. 128, Springer Verlag, Berlin (1991)
[28] Somers L.M.T., The Simulation of Flat Flames with Detailed and Reduced Chemical Models,
Ph.D. thesis, Eindhoven University of Technology (1994)
[29] Thiart G.D., Finite dierence scheme for the numerical solution of uid ow and heat transfer
problems on nonstaggerd grids, Num. Heat Transfer, Part B 17:4162 (1990)
[30] Turanyi T., Parameterization of reaction mechanisms using orthogonal polynomials, Comp.
Chem. 18:4554 (1994)
[31] Vreman A.W., Albrecht B.A., van Oijen J.A., de Goey L.P.H. and Bastiaans R.J.M., Premixed
and non-premixed generated manifolds in large-eddy simulation of Sandia ame D and F, Combust. Flame, Accepted, (2008)
[32] Warnatz J., Resolution of gas phase and surface combustion chemistry into elementary reactions,
Proc. Combust. Inst. 24:553579 (1992)
[33] Williams F.A., Combustion Theory, Addison-Wesley Publishing Company, Redwood city (1985)
[34] Williams F.A., Turbulent combustion, in The Mathematics of Combustion, edited by J. Buckmaster, pp. 97131, Society for Industrial and Applied Mathematics, Philadelphia (1985)
114
Part 3
Turbulent Combustion
115
Chapter 7
Introduction to turbulent combustion

D.J.E.M. Roekaerts
7.1
Introduction
Turbulent combustion is of interest both because of its practical importance and because the fundamental scientic questions it raises. Some practical systems in which turbulent combustion takes
place are furnaces, gas turbines, internal combustion engines and ares. The interest of society in an
economic, clean and safe operation of these devices is an important driving force for the development
of experimental techniques and models to acquire good understanding of turbulent combustion. Such
understanding is needed in the design of systems producing minimal environmentally harmful emissions. For example, the strong dependence of nitric oxide (NO x ) formation on temperature has made
it necessary to gain a better understanding of temperature uctuations and their inuence on mean
chemical reaction rates. Apart from this, the global environmental eect by combustion products
(CO2 ) provides a drive towards more ecient and integrated combustion processes using fuels with a
low C/H ratio.
The combination of non-linear uid dynamics with non-linear kinetics, causes very complex phenomena. Combustion in laminar ames is already complex due to the combination of complex diffusion properties and complex kinetics with a wide range of chemical time and length scales. But
turbulence adds to this the complexity of a wide range of time and length scales in ow phenomena and makes the need for eective simplied models even stronger. Here simplied in the rst
place refers to the fact that an incomplete and approximate picture of the phenomena is given. The
model itself may in fact have a quite complex appearance. Dierent models are developed for dierent types of ames, e.g. turbulent premixed gaseous ames, turbulent non-premixed gaseous ames
(also called turbulent diusion ames) and spray ames [1]. This chapter gives an overview of the
modeling approaches for the case of gaseous combustion.
The type of turbulent ame structure depends on the relative magnitude of the scales of turbulence
and the scales associated with a local ame structure and this can be taken into account in the modeling. The smallest scale in multicomponent inert turbulent ow depends on the Schmidt numbers
(characterising the relative strength of diusion of momentum and mass). In gas phase systems the
Schmidt number is of order unity and the smallest scale is the Kolmogorov scale. The size of turbulent
eddies ranges from the integral length scale t to the Kolmogorov scale K . The turbulence Reynolds
number is Ret = t u /, where u is the standard deviation of the velocity uctuations and the kinematic viscosity. The ratio of the integral time scale to the Kolmogorov time scale is tt /tK = Re1/2 .
t
Let tc be a time scale associated with the chemical conversion then the relative rate of ow and chemical processes is characterised by the dimensionless numbers Damk hler number Da = tt /tc and
o
117
118
Chapter 7 D.J.E.M. Roekaerts
Karlovitz number Ka = tc /tK . The Karlovitz number is related to Reynolds and Damk hler number
o
2 Ka2 . Small Da corresponds to slow reactions and large Da to fast reactions.
via Ret = Da
The Reynolds number characterises the extent of the turbulent energy cascade. The ratio of sizes
of large and small eddies scales as Re0.75 , the ratio of the time scale of large and small eddies scales as
t
Re0.5 . When Ka < 1, the laminar ame thickness of a premixed ame is smaller than the Kolmogorov
t
scale. To identify the smallest scale in non-premixed turbulent combustion it is necessary to consider
the variable describing the mixing of fuel and oxidiser (mixture fraction). The width of the reaction
zone in mixture fraction space and the spatial gradients of mixture fraction together determine the
width of the ame in physical space, which can be larger or smaller than the Kolmogorov scale.
7.2
Computational approaches
The physical and chemical phenomena in turbulent ames are described by the transport equations for
momentum, energy and mass fractions of all species, thermodynamic equations of state and expressions for kinetic rates and physical properties (viscosity, diusivity). Solving this system of equations,
resolving all details of the ow is called direct numerical simulation (DNS). This does not necessarily
imply that all features of combustion are resolved. DNS studies often use simplied physical properties and kinetic rates. The complexity of the ow increases with Re and Da and using DNS only
ows with suciently low Re and Da can be calculated. DNS is a useful tool to come to fundamental
understanding and also to produce data sets that can be used for the validation of models, but for the
prediction of the properties of turbulent ames in practical systems one has to resort to simplied
approaches.
Several general principles can be invoked to simplify the description of turbulent combustion.
Chemical processes which are faster than the fastest time scales of the ow relax to a quasisteady state at a given ow condition, and the steady state evolves in time at a rate determined by
the ow processes. By characterising the steady state via algebraic relations, or other projection
techniques, the details of the transient evolution to the steady state can be removed from the
model.
Due to the separation of time scales between on the one hand the processes determining local
ame structure (chemistry and laminar diusion) and on the other hand turbulent convective
processes, the local ame structures are quite universal. By characterising the possible ame
structures by a few independent variables a large simplication is achieved.
Fluctuating variables are characterised by statistical distributions. Often the characteristics of
the process can be captured by calculation only a few lower order moments of these distributions. By considering instead of the deterministic transport equations, an alternative set of
equations for the selected set of moments, a huge reduction in computational cost is obtained.
Because of the nonlinear nature of the Navier-Stokes equation and the chemical reactions this
reduction method is faced with essential diculties (closure problems).
The rst modeling principle is to select the most relevant thermo-chemical variables as independent scalars, for which full transport equations have to be solved, while keeping the other variables as
dependent scalars. For non-premixed ames the primary variable is the mixture fraction. In words it
can be dened as the fraction of the mass present locally, which originally is coming from the fuel
stream. Mathematically this can be dened as a normalised element mass fraction but other possibilities exist [60]. Several notations for mixture fraction, e.g. Z, and f , are used in the literature. For
119
premixed ames the primary variable is the progress variable. It characterises the evolution from cold
reactants to hot products. The most common notations for progress variable are c and r. Selecting
mixture fraction or progress variable as independent scalars, in both cases the proles of dependent
variables versus independent scalar are obtained from assumptions.
The second modeling principle is to identify on the basis of a time scale analysis the most prominent ame structures. It can be used to nd the best assumptions on the dependencies between independent and dependent scalars. E.g. in amelet modeling the dependencies are obtained from the
calculation of a laminar ame.
The third modeling principle is to invoke a statistical description. When considering models for
turbulent ows one can distinguish between traditional turbulence models calculating a set of mean
values (Reynolds Averaged Navier-Stokes model, RANS) and a more detailed approach calculating
all structures in the ow larger than a certain limiting scale (Large-eddy simulation model, LES). A
compact and precise introduction to LES models can be found in Ref. [47]. The development of LES
methods for reacting ows has been attracting growing attention in the last years. A textbook with
special emphasis on LES of combustion is [46].
In RANS studies one considers the turbulent ow from a statistical point of view and then restricts
the description to a subset of all statistical properties, e.g. in the simplest case only the mean value and
variance of velocity and concentrations. The mean values that are considered can be time-averages
or ensemble-averages depending on the situation. When not only average and standard deviation
but the complete set of statistical properties at a point in space is considered the method is called
Probability Density Function (PDF) method. In LES the evolution in time and space of a turbulent
ow is calculated in such a way that all ow details which are larger in space than a certain given
(lter) size are resolved. All smaller scales are only implicitly taken into account. In both classes
some properties are not explicitly calculated, in the rst class these are statistical properties such as
correlations between variables at dierent position in space, in the second class this concerns smallscale or subgrid motions. Therefore, in both classes some assumptions are needed before a closed set
of model equations is obtained: this situation is known as the turbulence closure problem. In the case
of RANS, chemical reaction terms represent a severe closure problem, because the mean chemical
reaction rate cannot be expressed as a function of mean temperature and mean concentrations. The
role of turbulent uctuations can not be neglected. Dierent solutions to this problem have been
proposed in the literature [18, 42, 47, 19], and will be reviewed below.
The computational expense of LES is smaller than that of DNS but can still be very high. In
principle one expects LES to be more accurate than RANS because a larger part of the turbulent
uctuations is resolved, rather than modelled. However, because chemical reaction is a local process
and in combustion, ame fronts can be thin, the construction of a good ltering procedure is a complex
problem. The resolved reaction rate, appearing in the Eulerian transport equation for the resolved
concentrations, cannot be expressed in terms of the resolved concentration and temperature elds.
The analogue in LES of the observation made in RANS that uctuations inuence the mean reaction
rate, is that in this case subgrid scale uctuations inuence the resolved reaction rate [50]. It can
be expected that the phenomena on the unresolved scale in LES will be less universal in reacting
ow than in non-reacting ow. At least they will be Damk hler number dependent. The essential
o
advantage of LES compared to RANS is that by grid renement (more precisely decrease of the lter
scale) more and more eects of the uctuations are resolved and less is modeled. E.g. in the case of
premixed combustion: more and more ame structures are computed and instantaneous zones of fresh
and burnt gases, where turbulence characteristics are quite dierent, are clearly identied [46, 56].
In the following, rst an introduction will be given to the statistical description of turbulence. To
start, instantaneous conservation equations are considered because they provide the basis for the aver-
120
aged equations and also for the transport equation satised by the probability density function. Next
some basic concepts from statistics are considered. Then the mean averaging and ltering procedures
used in RANS and LES methods are introduced. Finally a discussion of the main closure problems
and their possible solution in terms of second moment equations, PDFs, and fast chemistry models is
given. Further information can be found in the literature [15, 18, 42, 46, 48, 47, 58].
7.3
Instantaneous conservation equations
Conservation of mass is expressed by the continuity equation:

Ui
=0
+
t
xi
(7.1)
conservation of momentum (here in the absence of external forces) by the Navier-Stokes equation:
Ui +
U j Ui =
Ti j
t
x j
xi x j
(7.2)
in which p is pressure and T i j the viscous stress tensor. For a Newtonian uid the viscous stress tensor
contains only simple shear eects and can be written as:
T i j =
Ui U j
2 Uk
+
+
i j
x j
xi
3 xk
(7.3)
in which is the dynamic viscosity.

The conservation equations for species mass fractions, denoted here by the scalar vector , read:
+
U j =
J + S ()
t
x j
x j j
(7.4)
where J is the diusion ux and S is a chemical reaction source term. Neglecting eects of thermal
j
diusion and external body forces and assuming Fickian diusion, the ux reads:
J = D
j
x j
(7.5)
in which D is the mass diusion coecient which here for simplicity is assumed to be equal for all
species. The enthalpy conservation equation which in general is also needed, is also of the form 7.4
with the enthalpy source term in particular containing eects of radiative heat transfer. Enthalpy can
be included as component of the scalar vector. The description is completed with the thermodynamic
equation of state and the caloric equation of state (relating enthalpy and composition to temperature).
7.4
7.4.1
Basic denitions from statistics

One-point statistics
Consider the variable which is a function of space x and time t (denoted by (x, t)). The distribution
function F of is dened as:
F (; x, t) P{(x, t) < }
(7.6)
121
in which is the sample space variable of which takes all possible values of and P{...} stands for
the probability that the eld value at (x, t) is smaller than . The probability density function (PDF)
is dened as:
f (; x, t) =
F (; x, t)
(7.7)
The probability that a realization of (x, t) lies between 1 and 2 is given by:
2
P{1 < (x, t) < 2 } =
f (; x, t)dt = F (2 ; x, t) F (1 ; x, t)
(7.8)
The end values of F are appropriately dened as F() = 0 and F(+) = 1. Field values and
probability functions in general are considered to be functions of space and time but to simplify the
notation the variables x and t are often omitted without further notication.
The average or expectation of is denoted by E{}, , or by and can be expressed in terms of
the PDF by:
+
f ()d
(7.9)
In the same way, the expectation of any function Q of can be dened by:
+
Q =
Q() f ()d
(7.10)
Every sample of the variable can be decomposed into its mean and a uctuation according to:
= +
(7.11)
In turbulence modeling is called the Reynolds average

To clarify the notion of one-point statistics further, an example of two-point statistics is given.
Consider the two-point one-time statistics of the eld variable (x, t) dened by F (1 , 2 ; x, x + r, t)
which describes the probability that (x, t) < 1 , and (x + r, t) < 2 . This function cannot be
expressed in terms of the one-point statistics of . In the limit of |r| 0 this description reduces to
the combined statistics of and its gradients. This means that gradient statistics cannot be described
in terms of the one-point statistics of a variable and that such terms are unclosed in any one-point
closure.
7.4.2
Joint probabilities
We now consider combined statistics of several variables (e.g. velocities and species concentrations).
This provides information about correlations between variables.
Consider the n stochastic variables = 1 , 2 , ... n which have a joint probability function dened
by:
F () P{1 < 1 , 2 < 2 , ... n < n }
(7.12)
The joint PDF is dened by:
n
F ()
(7.13)
1 2 ... n
From the joint PDF of the statistics of one variable can be obtained by integration over the other
n 1 directions of the space:
f () =
f ( ) =
...
f () d1 ... d1 d+1 ... dn
(7.14)
122
This one-variable PDF is also called the marginal PDF of . As in the univariate case (see equation (7.10)) the expectation of any function of the variables i can be expressed as an integral over the
phase space of this function times the joint PDF.
7.4.3
Conditional probability
The conditional probability P{A|B} is dened as the probability that A occurs given that B occurs,
and is given by:
P{A, B}
P{A|B} =
(7.15)
P{B}
In the same manner the conditional PDF of 1 |2 can be dened as:
f1 |2 (1 |2 ) =
f1 ,2 (1 , 2 )
f2 (2 )
(7.16)
The expression f1 |2 (1 |2 )d2 now denes the PDF of 1 given the fact that 2 < 2 < 2 + d2 .
Conditional statistics are an important concept in PDF modeling. It turns out that all unclosed
terms in the exact transport equation for the joint scalar PDF can be written in terms of conditional
averages (e.g. of the uctuating velocity given a value of the scalar). These terms can in general not be
expressed as a pure function of the describing variables. These conditional averages Q(1 , 2 )|2 =
2 can be written in terms of the conditional PDF f1 |2 by:
Q(1 , 2 )|2 = 2 =
Q(1 , 2 ) f1 |2 (1 |2 )d1
(7.17)
7.4.4
Favre averaging
In turbulent ames, density can vary by a factor of ve or more and density uctuations can have
large eects on the turbulent ow eld. To simplify the equations describing variable density ow it
is common to use density-weighted (Favre) averaging.
Let the density be denoted as , and let the vector denote the thermo-chemical scalar variables
(e.g. species concentrations, temperature). In the low Mach-number limit, where it is assumed that
pressure variation does not aect the density, the density is a pure function of the scalar vector .
Favre averages are dened by:
Q
Q=
(7.18)
and Favre-decomposition into mean and uctuation is dened as:

Q= Q+Q
(7.19)
in which Q denotes the Favre-uctuation. Note that with the denition of Favre- and Reynolds
averages and uctuations:
Q =0
Q =0
and in general:
Q
It is also useful to dene the Favre-probability density function by:

f () = f ()
| =
()
= f ()
(7.20)
123
Because the density is a pure function of the scalar space variables, the conditional average reduces
to a function in and the second equality holds. Favre-averages can be expressed in terms of the
Favre-PDF according to:
Q=
Q() f ()d
(7.21)
The Favre-PDF has the same properties as a standard probability density function and all properties
discussed in the previous sections (multivariate statistics, conditional statistics) can be applied in a
straightforward manner. The Reynolds mean density can be computed as follows:
7.5
1
f ()d
()
(7.22)
Averaged equations in RANS
By averaging the ow equations a set of equations describing the mean ow properties is obtained.
Dening:
D
=
+ Ui
(7.23)
Dt t
xi
and Favre-averaging the momentum and species equations (7.2) and (7.4) gives:
D
p
Ui =
Ti j
u j ui
Dt
xi
x j
x j
(7.24)
=
Jj + S
u j
Dt
x j
x j
(7.25)
in which the last terms of the RHSs of the equations represent the Reynolds stress and the turbulent
scalar ux which occur in unclosed form. These terms contain second moments of the velocity distribution and joint velocity-scalar distribution respectively, and cannot be expressed in terms of the
rst moments or means. Another unclosed term in the equations is the averaged reaction term. Reaction rates are in general highly non-linear functions of composition and temperature and the averaged
reaction rate cannot be expressed as a function of mean concentrations.
The most common way to close the Reynolds stress is using two-equation turbulence models
(k- models), and the scalar ux is often closed using a gradient diusion assumption. They perform
reasonably for simple ows. In a more extended turbulence modeling approach transport equations are
solved for Reynolds stress and scalar ux (second moment closure [33, 35]). In an even more general
approach a modelled transport equation is solved for the one-point joint velocity-scalar probability
density function (PDF) [23].
7.6
LES equations
Conceptually, the formulation of a LES approach proceeds in four steps. First the denition of resolved variables (ltering) is given. Next the equations for the resolved variables are derived. Then
models are proposed for the unclosed terms and numerical solution procedures are chosen. In practice
the separation between errors induced by the choice of model and by the choice of numerical procedure are more dicult to separate in LES. Here we shall not go into details on numerical methods.
124
7.6.1

Filtering procedure
For systems with periodic boundary conditions, the ltering can be done in the spectral space, i.e.
on Fourier transformed elds, or in the physical space. In the study of combustion systems where
periodic boundary conditions are often not suited and ame structures are to be modeled, ltering in
physical space is most often used and we restrict to that case here.
The ltered and Favre-ltered values of the ow and thermochemical variables Q(x, t) are represented by Q(x, t) and Q(x, t) L = Q / , respectively.
The ltered quantity Q is dened as:
Q(x, t)
Q(x , t)G (x , x)dx
(7.26)
where G is the LES lter. refers to the lter width. The most popular lters are a box lter and a
Gaussian lter. In the rst case the lter corresponds to a spatial averaging of the quantity over a box
of size . In the second case the spatial averaging is with lter
G (x , x) =
6
2
3/2
exp
6
|x x|2 .
2
(7.27)
These lters are normalised, positive, and spatially and temporally invariant. Most often the mesh size
of the grid on which the resolved elds are obtained determines the lter width (implicit ltering)
but the mesh size could also be chosen dierently, see e.g. Ref. [2].
Any quantity Q may be split into a ltered part Q and an unresolved part Q , according to
Q = Q +Q . Two dierences between mathematical properties of the Reynolds averaging procedure
and the ltering procedure have to be taken into account when deriving the equations: rstly, ltering
the ltered eld a second time may lead to a dierent result than the ltered eld. Then Q may
be nonzero. Secondly, exchange of the order of ltering and derivation operators is valid only under
some restrictive conditions, e.g. it requires that the lter size is independent of spatial location. The
extra terms, arising by change of order of ltering and taking the derivative, are often assumed to be
handled as part of the subgrid scale models.
7.6.2
Filtered equations
By ltering the ow equations (without body forces) a set of equations describing the resolved ow
properties is obtained. Dening:
D
=
+ Ui L
(7.28)
Dt t
xi
and ltering the momentum and species equations gives:
D
Ui
Dt
Dt
Ti j
( U j Ui
xi
x j
x j

J +
x j j
( U j
x j
Uj
Uj
Ui L )
L )
(7.29)
(7.30)
in which the last terms at the right hand side of the equations represent the unresolved stress and
the unresolved ux which occur in unclosed form, as does the ltered chemical reaction rate. (Also
ltered values of laminar diusion uxes have to be modeled).

7.6.3
125
Modeling of unresolved turbulent uxes
The unresolved Reynolds stress can be closed by the following eddy viscosity assumption
( UU
with S
U L ) = 2 t S
1
(
2
U L) + (
(7.31)
the resolved strain rate tensor

S
U L )T
(7.32)
where t is the sub-grid kinematic eddy viscosity given by the Smagorinsky model as
t = (CS )2 |S|
(7.33)
Here is the lter width. The Smagorinsky constant CS for homogeneous and isotropic turbulence
has been estimated as CS 0.2, but it is not universal, and better models have been developed. E.g.
the dynamic model provides a methodology for determining an appropriate local value of CS [47].
The unresolved scalar ux can be closed using a gradient diusion assumption as:
U j
Uj
t L
Sct x j
(7.34)
where Sct is a subgrid scale turbulent Schmidt number. In RANS the assumption that the turbulent scalar ux is proportional to the gradient of the mean eld makes it impossible to predict the
experimentally observed phenomenon of counter-gradient transport. The LES assumption that the
unresolved scalar ux is proportional to the gradient of the resolved eld can still allow the prediction
the mean scalar ux is countergradient. This is because the mean scalar ux contains contributions
from both the resolved ux and the unresolved ux.
7.7
Closure of the chemical source term in RANS
Solving the moment conservation equations for turbulent reacting ows (7.25) the averaged reaction
rate term poses a serious problem. Because of the highly non-linear behavior of this term, the average
value cannot be expressed accurately as a function of the scalar mean, and information on higher
moments and correlations have to be included in the modeling.
In the presentation of closure models it is useful to distinguish between models which a priori
are generally applicable to all modes of combustion system (premixed, non-premixed or partiallypremixed) and models which are developed specically for one of these modes1 . Here generally
applicable means that the formulation of the model does not depend on the composition of the feed
streams and/or ame structure and the calculation proceeds in the same way for all modes of combustion. It should be remarked immediately that although applicable to any ame, the models need not
necessarily be accurate. And to increase the accuracy special subvariants may exist which are suitable
for specic modes.
In the case of fast reactions in a premixed situation the ame front separates regions with reactants
from regions with products. In the case of fast reactants in a non-premixed situation the ame fronts
arise where in the zones where the reactants meet after turbulent mixing. The physical mechanisms
1
A classication proposed by S.B. Pope et al. in a lecture presented at the 9th Int. Workshop on Measurements and
Computation of Turbulent Non-premixed Flames, Montreal, 2008.
126
active in premixed and non-premixed ames are essentially the same, but their relative importance is
dierent. In premixed ames ame propagation is essential and to describe it, a model for chemical
conversion at the ame front is needed, or at least a correlation for the ame speed. In non-premixed
combustion mixing is essential and to describe it a non-reacting scalar (mixture fraction) is sucient.
In both cases the inuence of turbulent uctuations has to be taken into account.
In the group of generally applicable models we list: laminar chemistry, PDF methods, eddy-break
up model, eddy dissipation concept model, linear eddy model. These shall be considered rst before
we turn to models for specic modes.
7.7.1
Models applicable to all modes of combustion
By denition the mean chemical source term is given by

S =
S () f ()d
(7.35)
and consequently the equation is closed provided we know the single point PDF.
Laminar chemistry
The simplest model for the mean chemical source term is to disregard the inuence of turbulent
uctuations and to evaluate the chemical source term by substituting mean mass fractions and mean
temperature in the laminar reactions rates. Mathematically this corresponds to the assumption that the
scalar PDF is a Dirac delta distribution concentrated at the mean value. This is an inaccurate model
in the case of fast reactions as prevailing in combustion, where uctuations play an important role.
Joint scalar PDF models
In joint scalar PDF models the source term is closed by explicitly calculating the PDF as part of the
model.
In the transported PDF approach a balance equation for the joint PDF of scalars is solved. The
essential advantage is that the PDF transport equation does not have a turbulence closure problem related to chemical reaction. Detailed kinetics can be included in exact form, but in practice
leads to high computational cost. In principle these methods are very general, but they are not
without closure problems because an equation for the one-point PDF contains unclosed terms
describing the eects of spatial gradients of the scalar variables (small scale mixing processes,
micromixing). Especially in the application to premixed combustion it is a factor limiting the
accuracy of the method. Standard micromixing models incorporate turbulent mixing time scales
but do not incorporate the laminar diusive mixing which becomes important in the amelet
regime. Transported PDF methods can be rather computationally expensive and use special
solution methods (stochastic simulation). In recent years PDF simulations with large detailed
mechanisms for the simplest fuels (hydrogen, methane) have been made (See e.g. [20]). The
drawback of computational cost has become less severe with the advent of ecient tabulation
algorithms and parallel computing algorithms (See [34] and references therein).
In presumed Probability Density Function (PDF) models a form of the PDF is assumed and is
quantitatively given as function of lower moments (mean, variance, co-variance). This success
of this method depends on whether sucient information is available to have a good presumed
shape of the PDF. For the case of many scalar variables this is usually not the case. Therefore
127
its application is limited to situations where the chemistry can be successfully reduced to just
a few independent variables. Because the availability of such a reduction may depend on the
mode of combustion, the details of this method can better be described in the frame of models
applicable to specic modes. It is recommended to select the independent scalars not only on
chemical criteria, but to also take into account available knowledge on joint statistical properties. In reaction progress variable models this is done by working with normalised scalars,
taking values in the interval [0,1], which are expected to be nearly statistically independent.
Joint velocity-scalar PDF models
In velocity-scalar PDF models one solves a transport equation for the joint PDF of velocity and
scalars. Knowledge of the joint velocity-scalar PDF at once implies a closure of all unknown
terms in the mean transport equations: the Reynolds stress, the Reynolds ux and the mean
chemical source term. The equation is solved with the help of Lagrangian methods, solving evolution equations for properties of uid elements (position, velocity, temperature, chemical composition, etc.) in contrast to Eulerian methods which solve transport equations for mass, momentum, energy and species. Many of these evolution equations are stochastic ordinary dierential equations, for which special solution techniques have been developed [18, 23, 42, 48, 49].
The method has a clear advantage when the gradient diusion assumption for the Reynolds ux
is doubtful. In practice the joint-velocity scalar PDF most often has been implemented in a
hybrid approach, where some lower order velocity statistics are simultaneously computed from
a transport equation. Then consistency between two ways to describe velocity statistics has to
be addressed [38, 61].
Further extensions including statistics of a turbulent mixing frequency have been introduced [47].
For a recent review of the application of PDF models to turbulent reactive ow, see Ref. [23].
Eddy Break-Up (EBU) Model
In the EBU model the mean reaction rate is assumed to be proportional to a turbulent mixing frequency
and the concentration of the limiting reactant. This is meant for the situation that turbulent mixing is
rate controlling. It contains empirical constants, which have to be estimated for each application. It is
based on a global representation of the conversion process and is to be combined with simple one or
few step chemistry models only.
Eddy Dissipation Concept (EDC) Model
In the Eddy Dissipation Concept (EDC) model [16, 21] the mean reaction rate is written as the product of a relaxation term and a scaling factor. Assuming that chemical reaction takes place within a
thin conned reaction zone which is typically smaller than the size of the computational grid, the
computational cell is conceptually divided into two sub-zones: the reacting ne structure and the surrounding uid. All the homogeneous chemical reactions are assumed to occur in the ne structures
that are locally treated as adiabatic, isobaric, Perfectly Stirred Reactors (PSR).(In some implementations also a plug ow reactor (PFR) is assumed which is computationally more ecient, but in fact
a dierent model.) The size of the ne structure and the mean residence time of the uid within the
ne structure is assumed to be of the Kolmogorov scale. The current mean state is relaxing linearly
to the PSR solution. The ratio of Kolmogorov scale to the large scale determines the scaling factor in
the mean reaction rate. The Eddy Dissipation Concept Model has the advantage that it incorporates
128
the inuence of nite rate kinetics at computational cost which is quite moderate compared to more
advanced models as the transported PDF method. This advantage comes at the cost of a much less
accurate description of turbulent uctuations.
Linear Eddy Model
The linear eddy model introduces the concept of a linear segment moving in the turbulent ow. The
concentration along this linear segment evolves in time due to diusion along the line and due to
discontinuous re-arrangement events induced by a hypothetical three-dimensional eddy [29, 52]. The
advantage of this model is that the re-arrangements can be described to reect the complete Kolmogorov (or any other) spectrum of turbulent ow. The linear eddy model is rather computationally
expensive, because the coordinate along the linear eddy also acts as an eective extra dimension of
space (In each cell of the normal computational grid one or more lines are present).
7.7.2
Models for the premixed mode
For premixed ames a regime diagram exists, categorising the ame type as function of ratio of
turbulence velocity uctuation to laminar ame speed and ratio of turbulent integral length scale to
laminar ame thickness. Flamelet models have been developed for the case that the reactions zones
are thin compared to the turbulent eddies. The features to be modeled in this approach are the actual
ame area and the resulting propagation velocity of the turbulent ame brush. Indeed, turbulence
increases the area of the ame front (wrinkling) and also modies the ame front (stretch, curvature).
This modies the overall conversion rate compared to that of laminar ame propagation. Correlations
have been developed for the relation between the turbulence intensity and the conversion enhancement
factor. In many models for turbulent premixed combustion these correlations are used in the model.
At rst the substructure of the ame front is disregarded. But in a renement step, details of the
substructure of the ame front, if required can be obtained from a separate laminar ame calculation.
BML model
The Bray-Moss-Libby model assumes innitely fast chemical reaction and correspondingly a innitely thin ame front. The model predicts the mean and standard deviation of the location of the
ame front. Extensions of this model give the mean chemical reaction rate expressed in terms of mean
crossing frequency of ame fronts.
Level set method
In the G-equation approach (level set method) an equation is solved for mean and variance of a function G, which is the distance to the nearest ame front. The model uses a correlation for the turbulent
ame speed controlling the propagation of the ame surface in space
Progress variable model
In a simple progress variable model the ame structure of a premixed ame is described by one
independent variable, the progress variable (e.g. the normalized product concentration). The mean
source term of the progress variable is unclosed and closure can be achieved using an assumed shape
of the progress variable PDF. In more detailed progress variable models, more than one independent
progress variable is used, in order to better represent the ame structures.
129
Flame surface density model

In ame surface density models the mean chemical reaction rate of the progress variable is expressed
as product of reaction rate at the ame surface times mean surface density. The ame surface density
is obtained from an additional modeled transport equation.
7.7.3
Models for the non-premixed mode
In non-premixed combustion the key variable is mixture fraction. A large simplication is realised
if mixture fraction is the only independent variable determining the ame structure. But in many
applications for several reasons one or more additional variables are needed. The discussion of these
models has two parts: rst the description of the ame structure has to be given, next the statistical
properties of the independent scalar variables has to be determined [40, 42].
Fast chemistry models
For adiabatic ames, in the limit of fast reactions the mixture fraction together with a fast chemistry
model completely denes the state of the system. The relation between the physical scalar variables
and mixture fraction is given by the specic conserved-scalar chemistry model used (e.g. mixed-isburnt model, equilibrium model, constrained equilibrium model, amelet model).
Mixed-is-burnt model. It assumes an innitely-fast irreversible global reaction of fuel and oxidizer to products which results in piecewise linear relations between composition and mixture
fraction. This model however does not include the formation of intermediate species like, in the
case of methane combustion, CO and H2 or radical species like O, H, and OH.
Chemical equilibrium model It does consider the intermediate species, but assumes that the
reactions are always fast enough to reach full chemical equilibrium, only depending on the
available atoms and energy. For the case of methane-air combustion this assumption is valid in
the high temperature regions of the ame but in the low temperature regions on the rich side
of the ame the slow burn-out of CO does not reach equilibrium and the CO concentration
is underpredicted. This eect can be remedied by making some extra assumptions, leading
to partial-equilibrium or constrained-equilibrium models. In these models some variables are
constrained to stay away from their equilibrium value. In the simplest case this constraint is
xed, in a more general case the constrained is an independent scalar variable.
Laminar amelet model. The local state of the mixture in the ame front is assumed to be
the same as that in a steady laminar diusion ame. Then by construction the balance between
diusion processes and nite rate of reaction, resulting in deviations from chemical equilibrium,
is taken into account. Apart from mixture fraction, a second variable, scalar dissipation rate or
strain rate is used to quantify the strength of the disturbance of the ame front by the ow.
Unsteady laminar amelet model. The unsteady laminar amelet model does not assume that
a balance has been reached between reaction and laminar diusion in the ame structure, but
rather solves the transient amelet equations for a the case of a time-dependent scalar dissipation
rate [44]. Since an unsteady amelet can only be calculated if the evolution of scalar dissipation
rate in time is known, it depends on information that comes available in the turbulent reacting
ow calculation. Whether or not an unsteady amelet model is needed, rather than a steady
amelet depends on the estimated turbulent time scale variation while moving with the ow.
130
In the case of non-adiabatic ames specic enthalpy has to be considered as independent variable in
addition to mixture fraction. The most important source term in the enthalpy equation often is the
radiative source term. The mean radiative source term is also unclosed. The closure of the emission
part of the source term is of similar nature of the chemical source term in the sense that it is fully
determined by the one point joint scalar PDF, but the closure of the absorption part is more complex,
because also depends on spatial correlations. For more information on the closure of turbulenceradiation interaction we refer to [8].
The thermo-chemical variables are non-linear functions of mixture fraction. Therefore to obtain
their mean value (e.g. mean density, mean temperature) it is not sucient to know the mean mixture
fraction. Characteristics of turbulent mixture fraction uctuations, characterised by its one point PDF,
have to be known. Many studies in the literature use an assumed shape of the mixture fraction PDF
in the form of a -function. The -function is selected because it can, as a function of its parameters,
take various forms that resemble physically realistic scalar PDFs (e.g. single-delta function PDFs
in fuel or oxidizer streams, or Gaussian-like PDFs in well mixed situations). The assumed shape of
the mixture fraction PDF depends on the mean and the variance, which are computed from modeled
transport equations. In these equations closure of the turbulent scalar ux is needed, which can be
done either by gradient diusion assumption or by solving a modeled scalar ux equation, which can
be expected to be more accurate. Some studies calculate the PDF of mixture fraction from a transport
equation, which in principle is more general.
In the case of non-adiabatic ames, the joint PDF of mixture fraction and enthalpy is needed. In a
rst approach the uctuations in enthalpy are assumed strictly correlated to the uctuations in mixture
fraction according to a prescribed relation [40].
The description of the main ame structure by fast chemistry models can be extended with transport equations for additional reacting variables for the slower secondary processes (NOx, soot). The
source term for these slow variables is unclosed but in some circumstances only uctuations of mixture fraction are important and the mean of the source terms of the additional variables can be obtained
by integration over mixture fraction uctuations only.
Mixedness-reactedness models
To describe the phenomenon that the states predicted by the fast chemistry models are not reached in
the turbulent ame, one or more progress variables can be introduced describing the evolution towards
the fast chemistry limit. One then has at least two independent variables: mixture fraction and progress
variable. The progress variable and the source term for the progress variable can be determined by
a variety of techniques exploiting analysis of the time scales determining the chemical process (CSP,
ILDM), or the chemical process coupled to laminar diusion processes (FPI, FGM, REDIM) [12, 51].
The determination of the joint PDF of the mixture fraction and the progress variables appears as a
modeling problem. FGM was initially developed for the premixed mode [39], with only reacting
progress variables, but later extended for application to other modes.
Conditional moment closure (CMC)
CMC is based on the observation that the most important uctuations to be taken into account in
non-premixed ames are the uctuations of mixture fraction. In this perspective the focus is on the
conditional expectation values with mixture fraction as the conditioning variable. In CMC modeled
transport equations are solved for the conditional moments of species mass fractions and temperature.
It turns out to be easier to close the mean source term in the conditional moment equation than the
mean source term in the standard moment equation. But the drawback is that the conditional moment
131
depends on an extra independent coordinate, namely the value of mixture fraction held xed by the
condition. This adds an extra dimension to the problem. In a three-dimensional ow problem the
conditional moment then depends on four coordinates. A complete description of the method can
be found in Klimenko and Bilger [31]. The model can also be formulated with other variables than
mixture fraction taken as conditioning variable, e.g. the progress variable in premixed combustion,
but is listed here as a model for non-premixed combustion since it is in this area that it has been most
often used. The model can be applied in combination with a detailed chemical mechanism. To obtain
unconditional mean values from conditional mean values, the PDF of the conditioning variable is
needed.
Multiple Mapping Conditioning (MMC)
The MMC model proposed by Klimenko and Pope [32] is a general model based on a combination
of PDF methods and CMC. It extends the application of the mapping closure which was introduced
earlier in the context of micromixing modeling in PDF methods. It has been applied to simulation of
turbulent non-premixed jet ames in ref. [57].
7.7.4
Models for the partially premixed mode
Real systems often are neither perfectly premixed nor perfectly non- premixed. For example: when
ignition in a nominally premixed combustion systems takes place after a time too short before a
prefectly premixed state is reached (stratication in an IC engine), or when in the region close to
the nozzle of a non premixed system turbulence levels are so high that ignition is suppressed and
mixing proceeds without reaction until a region with lower turbulence level is reached (lifted ame).
The partially premixed systems show some interesting new features. E.g. a triple ame structure
can occur at the lower edge of lifted non-premixed ame, which creates a need for appropriate model
extensions [46]. In the literature combinations of models for premixed and non premixed situations are
developed to cover the case of partially premixed reactants. For example: a combination of amelet
model and G-equation for the lifted ame. Or more generally, various combinations of models for
mixture fraction and models for progress variable.
7.8
Closure of the chemical source term in LES
Several methods have been proposed to account for the representation of subgrid-scale mixing and
chemical reaction in a LES. Because of the large formal similarity between modeling problems in
RANS and in LES, not surprisingly many closures for LES of reacting ows are direct analogues
of closures in the context of RANS. E.g. the analogue of the probability density function (PDF) of
uctuations in RANS, in the context of LES is a probability density function representing the relative
occurrence of values of physical quantities on the sub grid scale. Such a density in the literature has
been called large-eddy PDF (LEPDF) or also ltered density function (FDF), referring to the LES
approach of calculating only phenomena at scales larger than a given lter scale [50, 23]. Here we
shall use the name FDF. The ltered source term is known if the FDF is known and the FDF satises
similar equations as the PDF. In premixed combustion, the displacement speed of the resolved ame
front in LES appears in the same way in model equations as the turbulent ame velocity in the RANS
equations. Reviews have been published by Vervisch and Veynante [56], Janicka and Sadiki [27] and
Pitsch [45]. Here we briey mention some of the approaches proposed in the literature, using the
same subdivision as used above for RANS.
132
7.8.1

Models applicable to all modes
Filtered Density Function method

In the combination of LES and FDF, phenomena on scales larger than the lter scale are explicitly
calculated and the phenomena on smaller scales are represented in a statistical sense using the ltered
density function. The FDF can be obtained by assumed shape methods or by Monte Carlo methods in
a similar way as the PDF. Jones et al. (See [28] and references therein) have developed and applied an
alternative solution method, using the representation of the FDF by a number of continuous stochastic
elds. Recently also also the joint scalar-velocity-frequency FDF method has been developed [54].
Linear Eddy Model
In the application of the linear eddy model (LEM) to LES ([6, 30, 7] and references therein) phenomena at all length and time scales relevant at the subgrid scale are included, but in one dimension only.
Turbulent mixing is modeled by a stochastic rearrangement of subsets of the one-dimensional prole
of the scalar elds. One-dimensional laminar diusion, and possibly reaction, is included explicitly.
A collection of linear eddies represents the ow at the subgrid level. The linear eddies move according to both resolved and unresolved, hence modeled, ow eld. The linear eddies also interact with
each other at the subgrid scale. Recently the linear eddy model has been combined with tabulated
chemistry methods [53].
Articially thickened ame
By rescaling the laminar diusion coecients with a factor F > 1 and simultaneously rescaling
the reaction rate with a factor 1/F, the ame system is transformed into another system with, the
same laminar ame speed (in case of premixed combustion), but with a larger ame thickness [9].
The transformed system can be better resolved using a certain lter size and the objective is that the
macroscopic ame properties are the same as that of the original unscaled system. However, since the
Damk hler number of the articially thickened ame is also smaller by a factor F than the original
o
ame the turbulence-chemistry interaction is not invariant under the transformation. To accommodate
this the eciency function has been developed which describes the necessary correction of the reaction
rate to accommodate for that eect. The input parameters in the eciency parameter are related to the
subgrid scale turbulence. So, although the scaling principle underlying the articially thickened ame
model is applicable to all modes of combustion, the practical determination of an accurate eciency
function is specic for premixed combustion. The model recently has been applied to LES of swirl
stabilised premixed turbulent ames of interest for gas turbine combustion [11].
7.8.2
Models for the non-premixed mode
Flamelet models
Exploiting that combustion is a fast process, non-premixed combustion is described by solving the
transport of a conserved scalar (mixture fraction) and obtaining the chemical composition and temperature from a fast chemistry model. Assumed FDF models relate the subgrid distribution of mixture
fraction to its low order statistical moments. The mean is known from the resolved eld, but the
variance must be modeled. Dierent approaches for modelling the subgrid variance have been used.
Pitsch [43] has used unsteady amelet modeling to calculate a methane/air diusion ame (Sandia
Flame D). The method has been extended to include radiative loss and prediction of NOx-formation
[24].
133
Mixedness-reactedness models
Vreman et al [59] have applied the amelet generated manifold in LES of the Sandia piloted jet ames.
They compare the performance of FGMs based on non-premixed and premixed manifolds.
Conditional moment closure
CMC has rst been applied to LES in Ref. [5]. In the application to LES, apart from the subgrid FDF
of mixture fraction, also conditionally ltered value of the dependent variables) are computed. For an
application to a lifted ame, see [37]. Details on implementation issues can be found in Ref. [55].
7.8.3
Models for the premixed mode
LES of premixed combustion in general has to cope with the situation that the ame thickness L of
a premixed ame is smaller than the LES lter size . Therefore the ame front is not resolved in
the calculation, and step changes over one grid cell occur in resolved variables. To overcome this
diculty specic approaches, sometimes closely related to each other, have been proposed [9].
Turbulent ame speed closure
Several closure models use the ame speed. E.g. in the levelset method, the key idea is to track the
position of the ame front using a eld variable G. The ame surface is associated to a given iso-level
G = G . The G-eld does not have to follow the gradients of the progress variable (e.g. temperature)
and can be smoothed out to be resolved on a LES grid. The challenge is to propose a model for
the ame speed of the resolved front, appearing in the ltered G-equation. Fiorina et al [17] have
developed LES approaches using tabulated chemistry with the specic objective to recover the correct
laminar ame propagation speed of the ltered ame front when subgrid scale turbulence vanishes as
LES should tend toward DNS in that limit.
Flame surface density method
Another approach, developed in ref. [2] starts from the balance equation for the progress variable.
The ltered equation is solved on a mesh, but the mesh size is chosen ner than the lter width.
In this way the computational eciency of LES can be combined with resolution of the gradients
in the resolved progress variable. A ame surface density formulation is proposed to describe the
subgrid contribution of molecular diusive and reaction rate processes. The quality of this closure has
been investigated by comparing with data from a DNS. For a recent application of an algebraic ame
surface density equation in LES of ame propagation around obstacles, see [22].
7.8.4
Models for partially premixed mode
Domingo, Vervisch and Bray [13] have developed LES models for partially premixed ames in nonpremixed turbulent combustion. (I.e. when fuel and oxidizer have mixed before burning). Combustion in the premixed mode is assumed to occur in the thin amelet burning regime. Subgrid closure
schemes for fully premixed and non-premixed combustion are combined. A criterion using the resolved unsteady fuel and oxidiser elds is introduced to discriminate between the various degrees
of partial premixing and to estimate the proportions of premixed and non-premixed combustion at
subgrid level.
134
In the prediction of ame stabilisation of a lifted ame in hot vitiated coow several new variants of models combining features from premixed combustion, non-premixed combustion and autoignition have been proposed. E.g. Domingo et al [14] have used a chemistry model combining
premixed ame propagation and auto-ignition. The same ame was simulated by Ihme and See
[25]. Their model employs an unsteady non-premixed amelet formulation to describe the transient evolution of all thermochemical quantities during the ame ignition process. The quantities are
parametrised by mixture fraction, reaction progress parameter and stochiometric scalar dissipation
rate. A presumed PDF is used for the subgrid uctuations in mixture fraction and reaction progress
variable.
7.9
Conclusion
While preparing model calculations of turbulent reacting ow it is always good to check the mode of
combustion and the value of the relevant Damk hler numbers. This allows to identify the turbulenceo
chemistry interaction regime and to choose the most suitable model accordingly.
RANS modeling of turbulent combustion faces a severe closure problem of the mean chemical source term and dierent closure methods have been developed. RANS/amelet models and
RANS/PDF models are more appropriate respectively for situations with fast reactions and slow reactions compared to the ow. However, it cannot be expected that there is a sharp distinction between
amelet and non-amelet regimes.
Due to the growth in available computational power, accurate LES of labscale turbulent ames
has become feasible. In most applications the thickness of the ame zone is smaller than the lter
width and also LES faces the problem of the closure of the chemical source term. Because a part
of the turbulent spectrum is resolved and the modeling assumptions only concern the part of the
spectrum that is not resolved, the predictions using LES can be expected to be more accurate then
those using RANS. Further development of LES models for turbulent combustion currently is a very
active area of research. Nevertheless also RANS-based models remain of great practical use because
of the availability of well-tested software, a signicantly lower computational cost and the presence
of a knowledge base from the large range of applications already made.
References
[1] R.A. Bilger, S.B. Pope, K.N.C. Bray, and J.F. Driscoll. Paradigms in turbulent combustion
research. Proc. Combust. Inst., 30: 21-42, 2005
[2] M. Boger, D. Veynante, H. Boughanem, and A. Trouv . Direct numerical simulation analysis of
e
ame surface density concept for large eddy simulation of turbulent premixed combustion. In
Twenty-Seventh Symposium (International) on Combustion, 1998.
[3] N. Branley and W.P. Jones. Large eddy simulation of a turbulent nonpremixed ame. Combustion and Flame, 127:1913-1934, 2001.
[4] S.M. de Bruyn Kops and J.J. Riley. Mixing models for large-eddy simulation of nonpremixed
turbulent combustion. J. Fluids Eng, 123:341-346, 2001
[5] W.Kendal Bushe and Helfried Steiner. Conditional moment closure for large eddy simulation of
nonpremixed turbulent reacting ows. The Physics of Fluids, 11:18961906, 1999.
135
[6] William H. Calhoon Jr. and Suresh Menon. Linear-eddy subgrid model for reacting large-eddy
simulations: heat release eects. In 35th Aerosciences meeting and exhibit, 1997. AIAA 970368.
[7] V.K. Chakravarthy and S. Menon. Linear eddy simulations of Reynolds number and Schmidt
number eects of turbulent scalar mixing. The Physics of Fluids, 13:488499, 2001.
[8] P.J. Coelho. Numerical simulation of the interaction between turbulence and radiation in reactive
ows. Progress in Energy and Combustion Science 33:311383, 2007
[9] O. Colin, F. Ducros, D. Veynante, and T. Poinsot. A thickened ame model for large eddy
simulations of turbulent premixed combustion. The Physics of Fluids, 12:18431863, 2000.
[10] P.J. Colucci, F.A. Jaberi, P. Givi, and S.B. Pope. Filtered density function for large eddy simulation of turbulent reacting ows. The Physics of Fluids, 10:499515, 1998.
[11] A. De and S. Acharya.
Large eddy simulation of a premixed bunsen ame using
a modied thickened-ame model at two reynolds numbers. Combustion Science and
Technology,181:12311272, 2009.
[12] S. Delhaye, L.M.T. Somers, J.A. van Oijen, and L.P.H. de Goey. Incorporating unsteady oweects in amelet-generated manifolds. Combustion and Flame, 155, 133144, 2008
[13] Pascale Domingo, Luc Vervisch and Ken Bray. Partially premixed amelets in LES of nonpremixed turbulent combustion. Combust. Theory Modelling, 6:519-551, 2002
[14] P. Domingo, L. Vervisch and D.Veynante. Large-eddy simulation of a lifted methane jet ame
in a vitiated coow. Combustion and Flame 152:415432, 2008
[15] C. Dopazo. Recent developments in PDF methods. In P.A. Libby and F.A. Williams, editors,
Turbulent reacting ows, pages 375474. Academic press, London, 1994.
[16] I.S. Ertesvag and B.F.Magnussen. The eddy dissipation turbulence energy cascade model. Combustion Science and Technology, 159:213235, 2000.
[17] B. Fiorina, R. Vicquelin, P. Auzillon, N. Darabiha, O. Gicquel, and D. Veynante. A ltered
tabulated chemistry model for LES of premixed combustion. Combustion and Flame, 157:465
475, 2010.
[18] R.O. Fox. Computational methods for turbulent reacting ows in the chemical process industry.
Revue de lInstitut Fran ais du P trole, 51(2):215243, 1996.
c
e
[19] R.O. Fox Computational models for turbulent reacting ows. Cambridge University Press,
Cambridge, UK, 2003
[20] K. Gkagkas and R.P. Lindstedt. Transported PDF modelling with detailed chemistry of pre- and
auto-ignition in CH4/air mixtures. Proc. Combust. Inst., 31:15591566, 2007.
[21] I.R. Gran and B.F. Magnussen. A numerical study of a blu-body stabilized diusion ame.
Part 2. Inuence of combustion modeling and nite-rate chemistry. Combustion Science and
Technology, 119:191217, 1996.
136
[22] S.R. Gubba, S.S. Ibrahim, W. Malalasekera, and A.R. Masri. An assessment of large eddy simulations of premixed ames propagating past repeated obstacles. Combust. Theory Modelling,
13:513540, 2009.
[23] D.C. Haworth Progress in probability density function methods for turbulent reacting ows.
Progress in Energy and Combustion Science, 36:168259, 2010.
[24] M. Ihme and H. Pitsch Modeling of radiation and nitric oxide formation in turbulent nonpremixed ames using a amelet/progress variable formulation. Physics of Fluids, 20:055110,
2008.
[25] M. Ihme
Prediction of autoignition in a lifted methane/air ame using an unsteady
amelet/progress variable model Combustion and Flame, Accepted for publication
[26] F.A. Jaberi, P.J. Colucci, S. James, P. Givi, and S.B. Pope. Filtered mass density function for
large-eddy simulation of turbulent reacting ows. Journal of Fluid Mechanics, 401:85121,
1999.
[27] J. Janicka and A. Sadiki Large eddy simulation of turbulent combustion systems Proc. Combust.
Inst., 30:537547, 2005.
[28] W.P. Jones and V.N. Prasad. Large Eddy Simulation of the Sandia Flame Series (D-F) using the
Eulerian stochastic eld method. Combustion and Flame 157:16211636, 2010
[29] A. Kerstein. Linear-eddy modeling of turbulent transport. Part 6 J. Fluid Mech., 231:361-394,
1991
[30] Won-Wook Kim and Suresh Menon. Numerical modeling of turbulent premixed ames in the
thin-reaction zones regime. Combustion Science and Technology, 2000.
[31] A.Y. Klimenko and R.W. Bilger. Conditional moment closure for turbulent combustion. Progress
in Energy and Combustion Science, 25:595687, 1999.
[32] A.Y. Klimenko and S.B. Pope. The modeling of turbulent reactive ows based on multiple
mapping conditioning. Phys. Fluids, 15:19071925, 2003.
[33] B.E. Launder, G.J. Reece, and W. Rodi. Progress in the development of a Reynolds-stress turbulence closure. Journal of Fluid Mechanics, 68:537566, 1975.
[34] L.Y. Lu, S.R. Lantz, Z.Y. Ren, and S.B. Pope Computationally ecient implementation of
combustion chemistry in parallel PDF calculations. J. Comput. Phys., 228:54905525, 2009.
[35] J.L. Lumley. Second-order modeling of turbulent ows. In W. Kollmann, editor, Prediction
Methods for Turbulent Flows, pages 131. Hemisphere Publishing Company, 1980.
[36] J.H. Mobus, P. Gerlinger, and D. Bruggemann. Comparison of Eulerian and Lagrangian Monte
Carlo PDF methods for turbulent diusion ames. Combustion and Flame, 124:519534, 2001.
[37] S. Navarro-Martinez and A. Kronenburg. LES-CMC simulations of a lifted methane ame. Proc.
Combust. Inst., 32:15091516, 2009.
137
[38] P.A. Nooren, H.A. Wouters, T.W.J. Peeters, D. Roekaerts, U. Maas, and D. Schmidt. Monte
Carlo PDF modeling of a turbulent natural-gas diusion ame Combust. Theory Modeling,
1:79-96, 1997
[39] J.A. van Oijen and L.P.H. de Goey. Modelling of premixed laminar ames using ameletgenerated manifolds. Combustion Science and Technology, 161, 113137, 2000.
[40] T.W.J. Peeters. Numerical modeling of turbulent natural-gas diusion ames PhD Thesis, Delft
University of Technology, 1995.
[41] N. Peters. Laminar diusion amelet models in non-premixed turbulent combustion. Progress
in Energy and Combustion Science, 10:319339, 1984.
[42] Norbert Peters. Turbulent Combustion. Cambridge University Press, Cambridge, 2000.
[43] H.Pitsch and H. Steiner. Large-eddy simulation of a turbulent methane/air diusion ame (Sandia ame D). The Physics of Fluids, 12:25412554, 2000.
[44] Heinz Pitsch. Unsteady amelet modeling of dierential diusion in turbulent jet diusion
ames. Combustion and Flame, 123:358374, 2000.
[45] H. Pitsch. Large-eddy simulation of turbulent combustion. Ann. Rev. Fluid Mech., 38:453482,
2006
[46] Thierry Poinsot and Denis Veynante Theoretical and Numerical Combustion. Second Edition,
Edwards Inc, 2005.
[47] Stephen B. Pope. Turbulent Flows. Cambridge University Press, 2000.
[48] S.B. Pope. PDF methods for turbulent reactive ows. Progress in Energy and Combustion
Science, 11:119192, 1985.
[49] S.B Pope. Lagrangian PDF methods for turbulent ows. Annual Review Fluid Mechanics,
26:2363, 1994.
[50] S.B. Pope. Computations of turbulent combustion: progress and challenges. In Proceedings of
the Combustion Institute, 23, pages 591612, 1990.
[51] D. Roekaerts, B. Merci, B. Naud, and U. Maas. Elimination of Fast Modes in the Coupled
Process of Chemistry and Diusion in Turbulent Nonpremixed Flames: An Application of the
REDIM Approach. Int. J. Multiscale Computational Engineering, 7:487508, 2009.
[52] Vaidynathan Sankaran and Suresh Menon Structure of premixed turbulent ames in the thin
reaction-zones regime Proc. Comb. Inst., 28:203-209, 2000
[53] B.A. Sen and S. Menon. Linear eddy mixing based tabulation and articial neural networks for
large eddy simulations of turbulent ames. Combustion and Flame 157:6274, 2010.
[54] M.R.H. Sheikhi, P. Givi, and S.B. Pope. Frequency-velocity-scalar ltered mass density function
for large eddy simulation of turbulent ows. Physics of Fluids, 21:075102, 2009.
[55] A. Triantafyllidis and E. Mastorakos. Implementation Issues of the Conditional Moment Closure
Model in Large Eddy Simulations. Flow, Turbulence and Combustion, 84: 481512, 2010.
138
[56] Luc Vervisch and Denis Veynante. Turbulent combustion modeling. In Lecture notes, Von
Karman Institute, 1999. 177 pages.
[57] K. Vogiatzaki, A. Kronenburg, M.J. Cleary, and J.H. Kent. Multiple mapping conditioning of
turbulent jet diusion ames. Proc. Combust. Inst., 32:16791685, 2009.
[58] J. Warnatz, U. Maas, and R.W. Dibble. Combustion. Springer-Verlag, Berlin, 1996.
[59] A.W. Vreman, B.A. Albrecht, J.A. van Oijen, L.P.H. de Goey and R.J.M. Bastiaans. Premixed
and nonpremixed amelet generated manifolds in large-eddy simulation of Sandia ame D and
F, Combustion and Flame, 153:394416, 2008.
[60] F.A. Williams. Combustion Theory. Benjamin/Cummings Publishing Co., Menlo Park, CA,
second edition, 1985.
[61] H.A. Wouters and P.A. Nooren and T.W.J. Peeters and D. Roekaerts Simulation of a blu-body
stabilized diusion ame using second-moment closure and Monte Carlo methods Proc. Comb.
Inst., 26, 177-185, 1996
Chapter 8
Non-premixed amelet methods

L.M.T. Somers
8.1
Flamelets: introduction
As for pre-mixed combustion, the modelling of turbulent diusion ames heavily relies on simplifying the chemical processes (and transport processes for that matter). Various approaches exist and
depending on the simplications dierent models emerge. A compact organization of such models
according to Vervisch and Veynante [1] or Poinsot and Veynante [2] is given by
Innitely fast chemistry (mixed is burnt)
Finite rate chemistry assuming a local diusive-reactive structure similar to the one observed in
laminar ames (amelet assumption)
Finite rate chemistry with a separated treatment of chemical reaction from molecular and heat
transport (Conditional Moment Closure (CMC), pdf-method). Diusion is then addressed using
turbulent micro-mixing modelling, while chemical sources can be dealt with in a exact closed
form (pdf).
For a denition of the dierent regimes in turbulent combustion modelling one may read the latest
book by Peters Turbulent Combustion [3]. Basic ingredients of the analysis is that the smallest
eddy-length-scale (viz. Kolmogorov scale) is larger than the thickness of the ame. Furthermore the
turn-over time of the smallest eddy should be longer than a typical reaction time. In a nutshell the
amelet treatment as covered in this lecture assumes that the ame is a thin interface between fuel and
oxidizer. In a more or less geometrical analysis the contributions normal to this surface are separated
from the ones along this surface. This leads to identication of the so-called amelets, which can be
associated with one-dimensional counterow laminar diusion ames (section 8.3).
8.2
Flamelet equations: derivation
Consider the previously introduced balance equations (2.12.3) and (2.26) of lecture 1 on pages 13
and 18. Assuming equal diusivities for fuel and oxidizer (Lefu = Leox ) one can perform a series
of manipulations with the balance equations for fuel and oxidizer to get rid o the chemical source
term. This yields what is known as a Schvab-Zeldovich equation [1, 2, 4, 5] for the so-called mixture
fraction Z:
Z
1
+ (vZ) =
Z
(8.1)
t
Le c p
139
140
Oxidizer
Figure 8.1: Denition of coordinate transformation.
Z=
Na
Fuel
The mixture fraction Z is dened according to

Z=
Yfu Yox + Yox,2

Yfu,1 + Yox,2
(8.2)
where Yox,2 is the oxygen mass-fraction in the oxidizer stream and Yfu,1 the fuel mass fraction in the
fuel stream (see gure (8.2)). The is dened by
=
ox Mox
.
fu Mfu
which is the stoichiometric oxygen-to-fuel mass-ratio according to the global reaction:

fu Fu + ox Ox Products.
This mixture fraction is used to perform a transformation of the mass-fraction and enthalpy equation
in a new frame of reference where Z is one of the co-ordinates (see gure 8.1). A local orthogonal
co-ordinate system attached to the surface of stoichiometry is introduced and the derivatives in this
plane are denoted by the subscript ,
Yi
+ v
t
(Yi ) =
2 Yi
+
2 Z 2
(D
Yi )
(ln(|
Z|))
Yi
+ i
(8.3)
In equation 8.3 the scalar dissipation rate is introduced:

Z
= 2D
x j
= 2D| Z|2
(8.4)
which is the inverse of a characteristic diusive time = 1 . As this time decreases mass and heat
transfer through the stoichiometric plane are enhanced.
If the amelet is thin, or equivalently if gradients perpendicular to the stoichiometric surface
Z = Zst are large compared to the derivatives (viz. -terms) along this surface, an order of magnitude
analysis1 , shows that the balance equations to leading order are given by:
Similar to that for boundary layer analysis
Yi
2 Yi
=
+ i
t
2 Z 2
(8.5)
141
and a similar equation for the enthalpy balance. This results in the observation that if unsteady effects are neglected2 , the diusion ame is fully determined by the mixture fraction Z and the scalar
dissipation rate , yielding
Yi = Yi (Z, ); T = T (Z, )
From the 3D-simulation of fully compressible time-dependent simulation of a mixing layer (gure 8.2) it is seen that ame is a thin interface separating oxidizer (top) and fuel (bottom). Large
gradients in Z exist perpendicular to the plane Z = Zst throughout the time-dependent simulation (gures 8.2c, d). Furthermore, a strong correlation is present between the chemical source term and the
stoichiometric plane. This seems to be a good example for the amelet regime.
In most implementations the amelet approach is used in such a way that the ame equations or
chemistry can be solved separately from the hydrodynamics. The which is a function of Z is then
replaced by its value at the stoichiometric plane st = (Zst ). These amelet-equations are solved
o-line, prior to the hydrodynamic simulation. As a function of the two parameters st and mixture
fraction Z a representative library is built (see gure 8.8 for an impression). During the hydrodynamic
simulation the library is used. The stiness due to the chemistry is in this way separated from the
ow computation. In the next section some general aspects of the amelet structure are discussed.
8.3
Laminar counterow diusion ame(lets)
The counterow geometry (gure 8.3) is often used in experimental as well as numerical studies. This
is mainly due to the fact that within certain approximations the problem description can be reduced
to a set of one-dimensional pdes. However, originally the opposed-jet conguration was studied by
Potter and Butler [7] to study fuels that were too reactive to be burned pre-mixed (e.g. rocket fuels).
Consider a two-dimensional counterow conguration (gure 8.3). It can be shown that there
exists a transformation in terms of a similarity coordinate that leads to a system of one-dimensional
dierential equations [8, 9]. This is a rather complicated procedure therefore here a simpler derivation
is presented [3]. For that we need to introduce the Ansatz
u = Ux
Ansatz.
Applying this Ansatz we get:

Continuity
Momentum
Mixture fraction
Mass fraction
v
+ U
y
U
v
y
Z
v
y
Yi
v
y
=0
= U 2 + P +
(8.6)
y y
Z
D
y
y
Yi
=
Di
+ i
y
y
=
(8.7)
(8.8)
(8.9)
Here denitions are according to the co-ordinate system dened in gure 8.3. The parameter P is the
axial pressure gradient and is related to the strain rate a
P = a2
2
This is supposed to be true. Only near to extinction unsteady eects play a role.
142
(a) Mixture-fraction. Time t1 .
(b) Mixture-fraction. Time t2 .
(c) Vectors Z, solid line Z = Zst at t1 .
(d) Vectors Z, solid line Z = Zst at t2 .
Figure 8.2: Results from a fully compressible time-dependent simulation [6] with a high-order dns-code. The
simulation is a 3D simulation of a mixing layer including combustion which develops a Kelvin-Helmholtz
instability at the interface. Pictures are at a certain early stage (a,c,e) of the simulation where vortex pairs are
starting to develop and at a later stage (b,d,f) where structure becomes less regular.
continued on next page. . .
143
(e) Chemical source term, solid line Z = Zst at t1 .
(f) Chemical source term, solid line Z = Zst at t2 .
Figure 8.2: Continued.
Oxidizer
Nozzle
flame
yv
xu
Stagnation Plane
Nozzle
Gaseous Fuel
Figure 8.3: Counterow diusion ame. Schematic of a practical set-up. Dashed line shows the stagnation
plane. The gray fat-line illustrates the ame.
144
The strain rate a is the applied velocity gradient at the boundary, is the viscosity.
Now two possible scenarios exist to get a well-dened system of equations. On one hand one can
impose two potential ows coming from opposite directions y = + and y = . This is true only
if the nozzles are far enough from the stagnation plane and the ame. On the other hand one could
impose no-slip boundary conditions at the location of the nozzles. However, this leads to an ill-posed
problem, having too much boundary conditions. For that scenario the strain-rate can no longer be
imposed and has to be an eigenvalue of the system. Here the rst scenario is considered. One can
now show that these equations lead to the amelet equations derived from the ame-normal analysis
in section 8.2.
Lets apply the transformation
Z
=
y y Z
to the equation for the mass-fractions above. We then obtain:
v
v
Yi Yi
Z Yi
=
D
+ D
y Z
y
y Z
y
Yi
Z
= D
Z
y
Yi
D
y y
y
2 Yi
+ i
Z 2
2 Yi
+ i
Z 2
Z
y
2 Yi
+ i
Z 2
0 = D
(8.10)
where it is now easily seen that this yield the amelet equations (equation 8.5, on page 140) already
introduced earlier.
For this one-dimensional system it can be shown [3] that within certain approximations one can
derive an analytical expression for as a function of the strain rate a and the mixture fraction Z. Using
separation of variables the solution to eq. 8.8 can be written as
Z = c1 I(y) + c2
where
y
v
1
exp
dy dy.
D
D
0
0
The constants c1,2 follow from appropriate boundary conditions Z = 1 for y and Z = 0 for
y . Of course this analysis does not help much per se since , v and D in the former equation
are not known at this point. However if a number of not to serious assumptions are made [3] the
following expression can be derived:
Z = 112 erfc / 2
(8.11)
I(y) =
with the non-dimensional co-ordinate3 :

=
a
D
y
0
dy
(8.12)
Apply the denition of the scalar dissipation rate and the error-function4 . Then invert eq. 8.11 to
express as a function of Z. This nally gives:
2
a
(y) = exp (2 erfc1 (2Z) )
(8.13)
3
4
The function erfc is the complementary error-function, erfc(x) = 1 erf(x) with erf the error-function.
x
Note that erf(x) = 0 exp(t2 ) dt. Thus erf(x)/x exp(x2 ).
145
erfc(x)
1.5
erf(x)
1
Figure 8.4: Error function erf and complementary

error function erfc. Note the denition erf(x) =
x
2( )1 0 exp(t2 ) dt.
0.5
0.5
1
10
10
thus connecting the amelet equations to the counterow diusion ame equations with a given
strain-rate a. The function erfc1 (x) denotes the inverse of the erfc(x)-function and not its reciprocal 1/ erfc(x). As an illustration in section 8.3.2 simulations of counterow diusion ames using
the set of equations (8.68.9). First however we will discuss the Burke-Schuman limit or equivalently
the innite rate chemistry.
8.3.1
Burke-Schumann limit
A particular solution of the amelet equations is named after Burke and Schumann. They applied
the limit of innite rate chemistry to the amelet equations. It can be seen that then an innitely thin
reaction layer exists at Z = Zst where all combustion takes place. Outside this layer the source term
vanishes and the amelet equation for the mass fractions and enthalpy reduces to a homogeneous one:
D
Z
y
2 Yi
=0
Z 2
with general solution Yi = c1 Z +c2 . This results in the following linear relations in composition space:
Z Zst
1 Zst
Z
= Yox,2 1
Zst
Yfu = Yfu,1
and Yfu = 0
for Z Zst
(8.14)
Yox
and Yox = 0
for Z Zst
(8.15)
In gure 8.5 the solution is presented. Due to the assumption of innite rate chemistry the BurkeSchuman solution is independent of the strain, therefore the solution is a function of Z alone.
8.3.2
Non-equilibrium chemistry
If the chemistry is not innitely fast the complete system of equations must be solved. Since chemistry is sti special computer codes [10, 11] are needed to solve the discretized equations (see also
lecture of Boonkkamp). The general package developed at the TU/e is tailored to model these kind
of equations. However, in contrast to packages used in turbulence modelling, here the original equations (8.6-8.9) are solved whilst other authors (e.g. Peters) solve the amelet-equations and assume
assume (Z) = (Zst ) as a constant having the value found at the stoichiometric plane. For the simulations presented here weve taken one specic, yet particular case. In all simulations the pressure is
146

Yfu1
Tst
T
Yfu
Figure 8.5: The Burke Schumann solution presented in mixture space
Yox2
Yox
Zst
4 bar and the temperature at the nozzle is 1400 K and the equivalence ratio5 is 1. The mechanism
used is presented in section 2.4 of lecture 2 on page 30.
From gure 8.6(a) one can see already that combustion takes place in a narrow region of the
simulation. To get optimal resolution yet keeping a reasonable amount of grid-points the computer
codes adapts the mesh to the solution. Although from gure 8.6a it can be concluded that fuel and
oxidizer only co-exist in a small region, in mixture fraction space gure 8.6b however the picture
is quite dierent. Especially in the high strain limit the oxidizer extends well into the fuel side as
does the fuel into the oxidizer side. For low strain limits only in a region around Z = Zst fuel and
oxidizer co-exist. In gures 8.6c,d important radicals are presented. In c) intermediates from the C1
reaction chain are given and in d) H and OH important members of the radical pool. The latter two
species are responsible for the propagation of the reaction chain (see lecture 2.4). It can be seen that
the intermediates like CH3 do not extend far into the oxidizer side. They are rapidly consumed at the
reaction layer. The radicals like OH on the other hand do not extend far into the fuel side. These are
rapidly consumed as the methane is attacked and converted to CO.
Figures 8.7 also show that with increasing strain the reaction zone becomes narrower, leading to
larger gradients in temperature accompanied by lower peak temperatures. Finally the heat-conduction
(and radical diusion) from the reaction zone is no longer balanced by the generation of heat by
combustion and the ame will extinguish. It is interesting to notice that the mixture fraction oers
a very nice co-ordinate transformation from a numerical point of view. All proles, including the
species, remain smooth nearly independent of the applied strain.
In an application of the amelet approach several ames in a certain range of conditions are
computed. The results are stored in a library as a function of the two controlling parameters Z and st .
Just to get an impression in the following picture the temperature and OH mole fraction is plotted as
function of the strain-rate a and Z.
In the nal section some aspects of the implementation in a simulation of turbulent combustion is
treated.
8.4
Presumed pdf modelling
At some point in modelling turbulent combustion one gets into trouble. If youre not using dns, but
apply some kind of averaging 6 , you run into problems with respect to unclosed terms. In addition
5
6
fu,1
= Yox,2
Either RANS (time-averaging) or LES (space-averaging).
147
0.2
0.2
(Mole fractions)
(Mole fractions)
O2
0.15
CH4
0.1
Xi
Xi
0.1
0.15
0.05
0.05
0
-1
-0.5
0.5
0.2
0.4
x (cm)
0.6
0.8
(a) Methane and Oxygen mass fractions vs. space. Complete domain.
(b) Methane and Oxygen mass fractions vs. mixture fraction. Nb. fuel side corresponds to Z > Zst .
0.006
0.01
Increasing
strain
OH
0.005
CH3
(Mole fractions)
(Mole fractions)
0.008
0.004
0.006
Xi
Xi
0.003
0.004
CH2O
0.002
0.002
0.001
-0.05
0.05
0.1
-0.1
-0.05
x (cm)
0.05
x (cm)
(c) H and OH radical. Domain (0.05, 0.1). The arrow

indicates increasing strain.
(d) Methyl and CH2 O. Domain (0.1, 0.05). The arrow

indicates increasing strain.
Figure 8.6: The mole fraction of several species at 4 dierent strain rates (a = 100, 1000, 10000, 20000 1/s) as
a function of x. The a = 20000 1/s is not converged and will extinguish. In gure (c) and (d) some characteristic
radicals are plotted.
2200
2200
Increasing
strain
T (K)
2000
T (K)
2000
1800
1800
1600
1600
1400
-0.1
1400
Z=Zst
-0.05
0.05
0.1
x (cm)
(a) Temperature vs. distance, zoomed: one tenth of the

complete domain.
0.2
0.4
0.6
0.8
(b) Temperature vs. mixture fraction.
Figure 8.7: Temperatures for 4 dierent strain rates (a = 100, 1000, 10000, 20000 1/s). The a = 20000 1/s is
not converged and will extinguish.
148

1
1
+03
2.2x10
+03
2.1x10
+03
2.1x10
+03
2.0x10
+03
1.9x10
+03
1.8x10
+03
1.7x10
+03
1.7x10
+03
1.6x10
+03
1.5x10
+03
1.4x10
0.8
0.7
0.6
0.5
-03
6.0x10
-03
5.4x10
-03
4.8x10
-03
4.2x10
-03
3.6x10
-03
3.0x10
-03
2.4x10
-03
1.8x10
-03
1.3x10
-04
6.5x10
-05
5.9x10
0.9
0.8
0.7
0.6
0.9
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
5000
10000
15000
20000
5000
10000
a [1/s]
15000
20000
a [1/s]
(a) Temperature
(b) OH radical
Figure 8.8: Illustration of two entries (T and OH) of a amelet library.
to closures applied in non-reacting turbulence modelling, in combustion extra closures are needed.
For instance the averaged equation for a species mass balance reads:
Yi
+
t
(vYi ) +
(vYi ) = i
(8.16)
where the contribution of molecular diusivity has been neglected since it is believed to be much
smaller than the turbulent diusion, (vYi ). As is common in turbulence modelling the this term is
modelled with the gradient ux approximation,
(vYi ) = DY Yi
(8.17)
with DY = T /Pr where T is the eddy viscosity and Pr a Prandtl-number of order unity. Here we
employed the usual Favre-averages denition (see Lecture Roekaerts). After this introduction of the
closure for the scalar ux, which is similar to the closures used normally to describe the Reynoldstresses, the only remaining unclosed terms is the chemical source term i . Obviously, due to the
non-linear dependence of the chemical kinetics to the concentration of species and especially the
temperature i (Yi , T ) i (Yi , T ). At this point Probability Density Functions (pdfs) come into play.
In general the mixture fraction Z and its variance Z 2 are computed. The equation for the mixture
fraction is given by [3, 9]
Z
+ (vZ) DZ Z = 0.
(8.18)
t
Here as usual gradient transport is used to model the transport term. Additionally an equation for the
variance is needed (see [3]),
Z
(vZ
) DZ Z
= 2DZ
also using gradient transport. The scalar dissipation rate = 2D| Z|2 is modelled according to
= c Z
k
(8.19)
149
6
Za=0.6 gamma=0.5
Za=0.6 gamma=10
Za=0.6 gamma=4
Za=0.8 gamma=10
5
P(Z)
Figure 8.9: Illustration for dierent values of ,

and .
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
This relation is derived from an integral time for a scalar Z = Z 2 / and the assumption that it is
proportional to a characteristic ow time = /k. Most of the discussion normally is on the value
of the proportionality constant c . Values of 2 3 are encountered. However, some authors [12, 13]
claim that a more elaborate model is needed for the average scalar dissipation rate for which then a
transport equation is formulated.
The expectation of a certain variable can now be computed using a probability density function
and the computed amelets. By denition we have
F(x, t) =
0
F(Z, )P(Z, : x, t)dZd
(8.20)
whereas the joint-pdf P(Z, : x, t) is normally given as P(Z : x, t)P( : x, t) which assumes statistical
independence. If the shape of the probability function is presumed and not computed through a pdftransport equation the approach is called pre-sumed pdf modelling. The shape of the P(Z) is by most
authors given as a beta-function depending on Z and the variance Z 2 ,
P(Z : x, t) =
Z 1 (1 Z)1
( + )
()()
(8.21)
is the gamma-function and the two parameters = Z and = (1 Z) with given by,
=
Z(1 Z)
Z
are related to the mean Z(x, t) and variance Z 2 (x, t). The shape of the beta-function is chosen because
it can mimic various functions. For instance if is large it approaches a gaussian distribution while
for < 0 a singularity develops at Z = 0 and for < 0 at Z = 1 (gure 8.9). Peters [3] argues that the
beta-function does not always provide enough exibility. For instance a distribution with a singularity
at Z = 0 and an additional local maximum at an intermediate value of Z can not be represented.
Peters [14] advocates a composite model for such shapes which appear in shear layers and jets. With
respect to the pdf of the scalar dissipation rate a log-normal distribution is used [9]:
P( : x, t) =
1
ln() (x, t)
exp 2
2 (x, t)
(x, t) 2
1
(8.22)
150
The function (x, t) and (x, t) are given by the rst and second moment of the scalar dissipation rate:
= exp( + 0.52 ) and
= 2 exp(2 ) 1
Normally a value of 2 = 2 is proposed. In that way 2 does not need to be modelled.

Applying the pre-sumed pdf to the chemical source term we get
i (x, t) =
i (Z, )P(Z : x, t)P( : x, t)dZd
(8.23)
which nalizes the closure-problem. O course in literature a lot of discussion is about applicability
of the amelet concept. Currently time-dependent eects seem to be the center of attention as can
be read in recent issues of Combustion and Flame and proceedings of latest symposia of the Combustion Institute. More details of the amelet concept can be read in the excellent book Turbulent
Combustion of N. Peters [3], or Turbulent Reacting Flows edited by Libby and Williams [15]. Also
recommended for further reading are the Lecture notes of Vervisch and Veynante [1] and the book
by Poinsot and Veynante [2] as they give some nice similarities between modelling approaches for
pre-mixed and non pre-mixed turbulent combustion in general.
References
[1] L. Vervisch and D. Veynante, Introduction to Turbulent Combustion Modelling, Von Karman
Lecture series, VKI LS 199904, 1999.
[2] T. Poinsot, L. Vervisch and D. Veynante, Numerical Turbulent Combustion Modelling, 2002.
[3] Peters, N., Turbulent Combustion Cambridge Monographs on Mechanics, Cambridge University
Press, Cambridge UK, 2000.
[4] Williams, F.A. Combustion Theory, Second Edition, Addison-Wesley Publishing Company,
Redwood City 1985.
[5] Peters, N. and Williams, F.A., The asymptotic structure of Stoichiometric Methane-air ames,
Combustion and Flame 68, p. 185, 1987.
[6] Bastiaans, R.L.G.M., Somers, L.M.T. and de Lange, H.C., DNS of non-premixed combustion in
a compressible mixing layer., In: Geurts, B.J. (ed), Modern Simulation Strategies for Turbulent
Flow, To be published.
[7] Potter, E.A., Butler, J.N., Title unknown, ARS J. 29:54, (1959).
[8] Dixon-Lewis, G., David, T., Gaskell,P.H., Fukutani,S., Jinno,H., Miller, J.A., Kee, R.J., Smooke,
M.D., Peters, N., Eelsberg, E., Warnatz, J., and Behrendt, F. Calculation of the structure and
extinction limit of a methane-air counterow diusion ame in the forward stagnation region
of a porous cylinder, Twentieth Symposium (International) on Combustion The Combustion
Institute, Pittsburgh, p. 1893, 1985.
[9] Peters, N. and Kee, R.J., The computation of stretched laminar methane-air diusion ames
using a reduced four-step mechanism, Combustion and Flame 68, p. 17, 1987.
[10] The CHEMKIN-III Collection, http://www.ca.sandia.gov/chemkin/,http://www.reactiondesign.com/.
151
[11] CHEM1D: a computer code for solving one-dimensional ame equations Eindhoven University
of Technology, Eindhoven, The Netherlands. http://www.combustion.tue.nl/
[12] Pope, S.B., Consistent modelling of scalars in turbulent ows Phys. Fluids, 26, p. 404, 1983.
[13] Jones, W. P., and Musonge, P., Closure of the Reynolds stress and scalar ux equations Phys.
Fluids, 31, p. 3589, 1988.
[14] Eelsberg, E. and Peters, N., A composite model for the conserved scalar pdf, Combust. Flame
50, p. 351, 1983.
[15] Libby, P.A., Williams, F.A. (eds), Turbulent Reacting Flows, Combustion Treatise, Academic
Press, London UK, 1993.
152
Chapter 9
Premixed amelet methods

R.J.M. Bastiaans
9.1
9.1.1
Introduction
Turbulent premixed combustion
The most important application of turbulent premixed combustion is for stationary gas turbines. These
are the machines to generate power, which is supplied to the grid. Here the big advantage of using premixing is because of the high conversion ratio, low UHCs and thus high eciency. Furthermore there
is the advantage of low NOx emissions at lean conditions at which the temperature stays relatively
low. This is called dry low-NOx combustion. The disadvantage is the stability of the ame; this
is a consequence of the turbulent propagation velocity which is a property of the geometry, mixture
and ow structure. If the ow is not controlled well, ash-back and blow-o can easily occur. This
can potentially result in explosions and signicant production loss. Also thermo-acoustics plays a
role in which pressure uctuations from heat release and acoustics might result in amplication. With
the application of new fuels the stability of combustion and the turbulent consumption might be quite
dierent compared to natural gas combustion. In the latter case preferential diusion plays almost
no role because the diusivity of methane is almost equal to its heat conduction. With new fuels this
might not be the case anymore. Recent literature about turbulent premixed combustion modeling and
concepts behind it comprise Bilger et al. [2] and Driscoll, [9].
9.1.2
Requirements and methods
In turbulent combustion, the most straightforward way to solve the governing system of equations
numerically is by resolving all spatial and temporal derivatives that occur. This should generally be
performed for the density, an additional thermodynamic quantity, e.g. temperature, internal energy or
enthalpy, three velocity components and all the species describing the combustion process. We will
denote the number of species with n s , so that the total number of unknowns and equations is equal
to 5 + n s . The method to do so is called Direct Numerical Simulation, DNS. In fact the term DNS
originates from the community dealing with nonreactive ows. For combustion problems ideally
a detailed mechanism is used to perform DNS. Usually one thus arrives at a very sti system of
equations which needs a very high grid point density to resolve the inner structure of the ame with
many radicals and minor species. This requires a massive supercomputer to perform calculations
even for a very small physical domain. Therefore many of these exercises are performed in 2D,
but the rst 3D calculations appear in literature, [17, 31]. Note that turbulence is an essentially 3D
phenomenon, but also 2D DNS calculations are valuable. However using detailed chemistry in DNS
is not necessarily the case. Also cases with reduced chemistry (e.g. single step chemistry, amelet
153
154
Chapter 9 R.J.M. Bastiaans
methods) are called DNS, especially when the combustion is in the amelet region of the Borghi
diagram, as modied by Peters, [25]. In that case the hydrodynamic or aerodynamic turbulence is still
fully resolved.
If one wants to calculate turbulent ows without resolving everything, one has to model the inuence of the part that can not be resolved. Usually this is performed by applying and averaging
procedure to the equations. If a stationary turbulent ow is averaged in time or over an ensemble
of common phase, we obtain the Reynolds Averaged Navier Stokes equations or RANS. Thus one
obtains a steady solution and the inuence of all uctuations should be modeled. Another way of
dealing with the reduction of computational eort is to apply a local spatial average. This lters out
the smallest scales of motion, which are more homogeneous and isotropic. The inuence of these
scales is easier to model and less sensitive. RANS is often used in industry but this is changing to
application of LES with increase in computer power. DNS is used for validation of LES models.
9.1.3
Length and time scales
It is instructive to present an estimate on the computational grid needed to perform a DNS of a turbulent premixed combustion system. Let the large eddies of the velocity eld have characteristic length,
rms-velocity and time scales , u and = /u. Here the length scale is a large scale turbulent length
scale = k3/2 /, not a geometrical length scale [29], with k the turbulent kinetic energy and the
turbulent dissipation rate. The ratio of to the length scale of the smallest eddies (Kolmogorov
scale) is of order / Re3/4 , with Re = k2 /. To resolve all eddies the cell size must be of the
order of the scale of the smallest eddies and the size of the computational domain must be larger than
the scale of the large eddies. The ratio of these scales then determines N, the required number of
cells per direction in space. Since it is proportional to Re3/4 , the number of cells in three dimensions
scales as Re9/4 . Taking into account also the need to have an accurate solution in time one nds that
the computational eort scales with Re3 [29]. This is based on a CFL condition (with x = ), combined with a statistical viewpoint that the total integration time should be at least comparable to .
For compressible codes CFL timesteps are even more restricted resulting in an eort proportional to
Re3 /Ma.
The requirement to resolve all ame fronts is an independent requirement. Let the laminar ame
have thickness L and speed sL . The (diusive) ame thickness can be estimated by L /sL [27].
Assuming that the number of grid points needed to resolve a ame front is n (say 10), one has that the
new requirement on N is:
L
u
= x
=
=
(9.1)
N
n
nsL nsL Re
The variable denotes both viscosity and diusivity (The Lewis number was set to one). This leads
to
sL
N n Re
(9.2)
u
Using again L sL = and introducing the Damk hler number, the ratio of the the turbulence (mixing)
o
time scale, , and the ame time scale, tL = L /sL
Da =
/u
L /sL
(9.3)
one nds
N n Re Da
(9.4)
155
The Damk hler number expresses the combustion regime in the sense that for Da
o
1 we are in the
broken reaction zones regime and for Da
1 there is a amelet type of combustion. Depending on
the value of Da this is a weaker or stronger requirement than the requirement from turbulence only.
However the Dais more associated to non-premixed combustion in which the mixing time is relevant
to bring oxidizer and fuel together. For premixed amelet regime we have that Da
1 and the
requirement on the computational resources is stronger. Moreover additional stiness introduces by
the chemistry will also increase the computational time. In turbulent premixed (amelet) combustion
often the Karlovitz number is introduced, denoting the ratio of ame time to Kolmogorov time,
Ka =
t L 2
= L.
t
2
(9.5)
This number is a more natural indication of the physics for this case then Da. Application results in
N nRe3/4 Ka1/2 ,
(9.6)
which holds if the ame thickness is resolution limiting, i.e. for Ka < 1. One can also dene a
Karlovitz number based on the thickness of the reaction layer, also called the inner-layer, of the ame
instead of the entire structure. We will denote this number with Kail . The region enclosed between the
lines Ka = 1 and Kail = 1 is called the thin reaction zones regime; turbulent eddies can penetrate the
preheat zone, but they are not able to disrupt the reaction zone of the ame. Therefore, in the sense,
this region can be considered as belonging to the amelet regime, [16].
Introducing some representative values in the expressions above can be seen that the computational eort needed for DNS of practical combustion systems is still prohibitive.
9.2
DNS with amelet model
DNS is a very valuable tool, especially to validate turbulent combustion closure models for RANS and
LES. To that end a possible solution is to use amelet generated manifolds, FGM, to cover the chemistry, [20]. FGM is equal to a method called amelet prolongation of ILDM, FPI, [13]. Probably the
method was rst used (in a RANS context) in [4]. The idea behind the technique is that in premixed
systems, reaction progress is a key parameter and this parameter can be described by a combination of
chemical species representing the state of burning. In this way, premixed amelets are a natural choice
Figure 9.1: Structure of a lean turbulent premixed ame kernel. The isosurface is at the innerlayer of the ame,
coloured with the temperature. A cut-out is made to show the local ame structure.
156
Figure 9.2: Projection of ame paths (bold lines) on the plane z = 0 for a turbulent expanding ame at t = 0.2
ms. The thin lines are isotherms corresponding to the position of the unburned side, the inner layer and the
burned side of the ame. The vectors represent the gas velocity in the plane z = 0. Dimensions are in mm.
to build a database (manifold). Traditionally non-premixed amelets had been used to describe turbulent non-premixed combustion, using the scalar dissipation rate for covering non-equilibrium eects.
In order to describe turbulent mixing as well, the premixed amelets are used in a way in which both
mixture fraction and reaction progress were taken into account. Here the advantage is that a much
larger part of the reaction domain can be covered by premixed amelets, compared to non-premixed
amelets. Furthermore the method provides a physical framework in which more and more eects
can be taken into account. Practical and more theoretical studies were performed to investigate the
possibilities and limitations of the method.
Initial studies with FGM were focused on laminar combustion for premixed and partially premixed
cases, [20], [21]. The study of Eggels, [10], was a rst application to gas turbines. After successful
results obtained in these studies, the method was applied to 3D DNS for purely premixed cases,
[1, 23]. In 3D cases (9.1) the eect of ame stretch was studied on the local mass burning rates
by interrogating the entire ame structure by extracting local ame normal behavior, [24]. In these
studies the progress variable was taken as the normalized CO2 mass fraction. In these investigations it
was researched whether a single progress variable would be sucient to capture stretch eects. This
is explained in more detail in the next section.
In order to validate the method, 2D simulations have been carried out with detailed chemistry.
Here (initially) cylindrical ames were considered. It was found that the dierences between FGM
and detailed chemistry DNS in 2D, were very small. Normalized dierences were in the order of
102 to 103 for FGMs with one and two controlling variables respectively. These results are for lean
methane combustion in which all Lewis numbers were set to unity. Even at conditions of enlarged
turbulent uctuations, results were still very satisfactory. For non-unity but constant Lewis numbers
the structure with a 1D FGM was adequately captured but the dierences were a little larger with a
single progress variable.
157
Figure 9.3: Scaled mass burning ratemi/m0i at the inner layer as function of the Karlovitz integral Ka. Left:
low turbulence case, thin reaction zones regime. Right: high turbulence case, broken reaction zones regime.
Note that the Ka range for high turbulence is more than 10 times larger than for the low turbulence case.
9.2.1
Flame stretch
An important eect of turbulence applied to a ame is that the ame is stretched. Therefore the ame
is modulated and most importantly the burning speed is changed compared to unstretched conditions.
This can be analyzed by investigating the amelet equations, in which a local coordinate system is
used with a principle coordinate in the direction of the ame progress (e.g. temperature), as in [15].
In such a system stretch is explicitly taken into account. For the laminar burning velocity it follows
that
mb
= 1 Kab
(9.7)
m0
b
This is an exact equation for cases in which diusivity of fuel species equals the conduction, i.e. Lewis
number equals unity, Le = 1, which is the case for methane, and where the reaction zone is innitely
thin. In this expression we have the Karlovitz integral Kab dened as
Kab =
1
(m0 ) b
sb
KYds
(9.8)
su
in which is the local surface, K the local stretch rate and Y the progress variable (for more details
see [23] and [24]).
With the theory we can check the results as presented in the previous subsection. To that end
we have to track individual amelets in the turbulent eld. A visualization of such amelets is given
in gure 9.2. The results from the check can be observed in gure 9.3. Two cases are presented: a
case in which there is relatively low turbulence and a case with higher turbulence. It can be observed
that the FGM approach gives good results with a single progress variable for low turbulence and
even acceptable results for the higher turbulent case. The burning velocity is predicted very well and
therefore stretch eects are included in the right way.
158
9.3
LES modeling
In large-eddy simulation (LES) the larger three-dimensional unsteady turbulent motions are directly
represented (resolved), whereas the eects of the smaller scale motions are modeled (ltered). A
compact and precise introduction to LES models for cold ow simulations can be found in Ref. [29].
The development of LES methods for reacting ows has been attracting growing attention in the last
years. A recent textbook with special emphasis on LES of combustion is [27].
Developments have been made in dierent directions for dierent applications: atmospheric ows,
premixed combustion, non-premixed combustion. In this lecture an elementary introduction to LES of
reacting ow is given, followed by a short overview of some recent developments from the literature
and a presentation in some detail of one specic approach, the ltered mass density function (FMDF)
method.
The computational expense of LES is smaller but can still be too high. It will be of interest
when the most important phenomena to be studied are present in the large scale motions and when
the predictions are not sensitive to the model used to represent the unresolved motions. E.g. in the
study of combustion systems LES methods can be very relevant in the study of large scale combustion
instabilities. However, because chemical reaction is a local process and in combustion, ame fronts
can be thin, the construction of a good ltering procedure is a complex problem. The resolved reaction
rate, appearing in the Eulerian transport equation for the resolved concentrations, cannot be expressed
in terms of the resolved concentration and temperature elds. The analogue in LES of the observation
made in RANS that uctuations inuence the mean reaction rate, is that in this case subgrid scale
uctuations inuence the resolved reaction rate [28]. It can be expected that the phenomena on the
unresolved scale in LES will be less universal in reacting ow than in non-reacting ow. At least
they will be Damk hler number dependent. The essential advantage of LES, compared to RANS,
o
is that by grid renement (more precisely decrease of the lter scale) more and more eects of the
uctuations are resolved and less is modeled. E.g. in the case of premixed combustion: more and more
ame structures are computed and instantaneous zones of fresh and burnt gases, where turbulence
characteristics are quite dierent, are clearly identied [37, 27].
9.3.1
A reference approach with FGM
The governing equations

Here we will take the governing equations in which premixed as well as non-premixed combustion
can be taken into account. We will follow the approach of Vreman et al.[41],[40] in which also
experiences with dierent variants of the model are reported. It should be mentioned that the approach
is valid for methane ames in which the fuel has a diusivity that equals the heat conduction. In that
case no extra equations have to be taken into account. For other fuels preferential diusion might
become important, this is treated in section 9.4. Thus, in this case the Navier-Stokes equations with
159
parametrized chemistry read:

u j
+
t
x j
ui ui u j
+
t
x j
u j Z
Z
+
t
x j
Y u j Y
+
t
x j
= 0,
(9.9)
S i j
p
+2
,
xi
x j
Z
D
,
x j
x j
Y
D
+ Y ,
x j
x j
f1 (Z, Y),
f2 (Z, Y),
(9.14)
f3 (Z, Y),
(9.15)
(9.10)
(9.11)
(9.12)
(9.13)
where the summation convention over repeated indices is used, while , u, p, T , Z and Y represent
density, velocity vector, pressure, temperature, mixture fraction, progress variable, respectively.
The functions f j are used to dene the quantities that are retrieved from a amelet database. Since
all quantities in the amelet database can uniquely be tabulated as function of Z and Y, the entire
chemistry depends two variables only, Z and Y. The amelet database is a set of amelets, obtained
by solving the premixed amelet equations with detailed chemistry using the GRI 3.0 reaction scheme
[22].
The progress variable should be chosen such that its behavior on the amelet is monotonic. The
progress variable used in the partially premixed combustion case is a (linear) combination of species
produced by the combustion of methane
Y = YH2 /MH2 + YH2 O /MH2 O + YCO2 /MCO2 ,
(9.16)
where Yi and Mi represent the mass fraction and the molar mass of species i, respectively. For each
case a dierent denition of the progress variable can be applied in order to be able to describe the
kinetics better, depending on the basic structure of the combustion. Thus for the premixed Bunsen
ame, we again use another progress variable, the mass fraction of O2 . In a fully premixed case, the
mixture fraction is constant, such that then the equation for Z can be abandoned. For constant Z it is
convenient to use a scaled progress variable, c, varying between zero (unburned) and one (burned).
Thus in our premixed case, the equations (9.11) and (9.12) are replaced by the equation for c:
c u j c
c
+
=
Dc
+ c ,
t
x j
x j
x j
(9.17)
with c = Y . Then the functions f1 to f3 depend on c only.

It is remarked that T is almost reversely proportional to ; T is prescribed by the combustion
approximation, an approximate equation of state [22]. In addition the rate of strain is dened by
S ij =
1
2
ui u j 2 uk
+
i j ,
x j xi 3 xk
(9.18)
while the viscosity is a function of temperature according to Sutherlands three-coecient law. The
diusivity D is the diusivity of temperature,
/c p = 2.58 105 (T/298 K)0.69 ,
(9.19)
160
where is the thermal conductivity and c p the specic heat [34]. The diusivity Dc in the premixed
case is equal to D divided by the Lewis number of O2 (1.106). In the partially premixed case (ame
D) we use DZ = DY = D.
The ltered equations
Whereas DNS with amelet chemistry solves the unltered equations, LES solves the ltered equations. The basic lter notation in LES reads:
ui =
G(x, )ui d,
(9.20)
where is the ow domain and G is the lter kernel. Implementations in the present work use the
top-hat lter, dened by
G(x, ) = 1/3
if
|xi i | < i , i = 1, 2, 3;
(9.21)
and zero otherwise. Here denotes the lter width, equal to (1 2 3 ) 3 , and we use i = hi , where
hi is the local mesh-spacing in the xi -direction. In variable density ows it is convenient to use the
density-weighted or Favre lter as well, dened by ui = ui /.
The ltered equations are obtained by application of the Favre average to the equations in the previous subsection. The nonlinearities in the equations lead to unknown terms, which are either modeled, or neglected. There are four types of subgrid terms: subgrid terms arising from the nonlinearity
of the convective terms, subgrid terms arising from the nonlinearity of the chemical parametrization
(nonlinearities in f1 to f3 ), subgrid terms arising from the nonlinearity of the viscous terms, and subgrid terms due to the lack of commutation of lter and derivative on nonuniform grids. Types three
and four are neglected in the present work, type two will be considered in detail in section 3, and for
type 1 the eddy-viscosity/eddy-diusivity closure approach is adopted. As an example, we write out
the resulting equation used for c:
u
j c
c
t
c
+
=
(DY +
)
+ c ,
t
x j
x j
S ct x j
(9.22)
The last term is still in unclosed form (closures for this term will be presented in section 3). The
eddy-diusivity is the ratio of the eddy-viscosity and a constant Schmidt number of 0.4, according
to literature [26]. Unless mentioned otherwise, we use for the eddy-viscosity the model of Vreman,
[38]:
11 22 2 + 11 33 2 + 22 33 2 1/2
12
13
23
t = C
.
(9.23)
kl kl
Here the tensor equals the gradient model,
i j = 2 ki k j ,
k
ki =
i
u
,
xk
(9.24)
2
and the model constant is related to the Smagorinsky constant C = 2.5CS . In this work we take
CS = 0.1. A priori tests have shown that the gradient model of Leonard, [18], mimics the structure of
the exact turbulent stress quite accurately, Clark et al. [6]. Since straightforward use of the gradient
model causes stability problems, the gradient model was recast into the eddy-viscosity formulation
above. Model (9.23) has shown to be as accurate as the dynamic subgrid model [12] in wall-bounded
and free shear ow and, like the dynamic model and the wall-adapted local eddy-viscosity model [19],
161
but unlike the Smagorinsky model, the present model vanishes near walls and in transitional ow. In
fact it vanishes exactly for certain types of laminar ow, the same types of ow for which the exact
subgrid dissipation is zero [38].
In practice the beta-pdf approach is used to construct a table, where the ltered chemical quantities, depend on the ltered value, c, and its subgrid variance, var(c) = c2 c2 . The model for the
source-term reads
1
M =
c (c)Pc,var(c) (c)dc,
(9.25)
where P represents a probability density function of beta-shape with mean c and variance var(c).
While constructing the table, the beta-pdf integrals have to be calculated carefully [30] and need to be
checked for their convergence. The closure problem that remains is to provide a suitable model for
var(c). A similarity or gradient model is often adopted, e.g.,
a2 2 2
c
k
var(c)
(
) ,
12 xk
(9.26)
where a is assumed to be constant, or determined by a dynamic procedure. For smooth elds on the
scale of we can deduce a = 1 from a simple Taylor expansion. Next we derive an upperbound for
a. The beta-pdf requires c(1 c) 1 . Since we discretize c in Eq. (9.26) with second-order central
4
dierencing across a distance of 2, the absolute maximum of a rst-order derivative of c is 1/(2).
It follows that for a larger than 2, values larger than the theoretical beta-pdf maximum ( 1 ) may occur.
4
Results
The results of the above approach are summarized in Vreman et al. [41]. It appears that the method can
be used in premixed and non-premixed cases. In particular the eect of the subgrid scale contributions
was investigated for the premixed case, which was a 2D Bunsen ame. For non-premixed combustion
the inuence of the eddy viscosity model was reported. For this second case, in which a comparison
with experimental data was made, an LES was performed of a partially premixed jet diusion ame,
Sandia Flame D. Several subgrid models were compared, the Smagorinsky eddy-viscosity, where the
mean shear can be quite dominant, and three other models with less dependence on mean shear. The
latter three models produced results closer to the experimental data than the Smagorinsky model.
For the generality of our approach it was interesting to learn that the premixed LES strategy works
reasonably well, also for partially premixed ames.
A more extended research on the chemistry modeling in the case of Sandia ame D was reported
in [40]. Here it was found that the origin of the manifold, either constructed from premixed or nonpremixed ames, did not matter that much. This is illustrated in gure 9.4, which is taken from [40].
The result can be explained by the fact that the perturbations of the ame by turbulent stretching are
relatively low, that therefore the combustion is close to equilibrium, which is independent of the ame
structure, premixed or non-premixed.
An important issue is the interference of subgrid scale modeling with the discretization error. This
was studied in more detail in [39].
9.3.2
Dierent approaches
Several other methods have been proposed to account for subgrid-scale mixing and chemical reaction in a LES. Because of the large formal similarity between modeling problems in RANS and in
LES, not surprisingly many closures for LES of reacting ows are direct analogues of closures in the
162
0.02
0.015
0.015
0.01
0.02
0.01
0.06
0.08
0.06
0.04
0.04
0.005
0.005
0.02
0.02
0.2
0.4
0.6
0.8
0.2
0.4
0.6
0.8
(a)
(b)
Figure 9.4: Premixed (a) and non-premixed manifold (b) for the mass fraction of CO. The contour increment
equals 0.02. On top of the manifolds scatter plots obtained from the corresponding large-eddy simulations have
been drawn (ame D). The bold dashed line in (b) represents the non-premixed amelet for a = 1200/s; below
that line the manifold is theoretically undened.
context of RANS. An overview has been given by Vervisch and Veynante [37]. They conclude that
extensions of simple procedures such as neglecting uctuations of eddy break-up type of models were
not successful. Next they separately review approaches developed for non-premixed combustion and
for premixed combustion. Here we proceed similarly by refering to a selection of recent publications.
Only a qualitative description is given of the premixed approaches.
LES of premixed combustion in general has to cope with the situation that the ame thickness
L of a premixed ame is smaller than the LES lter size. Therefore the ame front is not resolved
in the calculation, and step changes over one grid cell occur in resolved variables. To overcome this
diculty specic approaches, sometimes closely related to each other, have been proposed ([7] and
references therein)
Artificially thickened flame By rescaling the ame thickness with a factor F > 1 and simultaneously rescaling the reaction rate with a factor 1/F, the premixed ame system is transformed into
another system with the same laminar ame speed, but with a larger ame thickness [7], which can
be described with less computational eort. However, since the Damk hler number of the articially
o
thickened ame is also smaller by a factor F than the original ame the turbulence-chemistry interaction is not invariant under the transformation. To accommodate this the eciency function has
been developed which describes the necessary correction of the reaction rate to accommodate for that
eect. The input parameters in the eciency parameter are related to the subgrid scale turbulence.
Flame-front tracking technique In the ame front tracking technique, the ame surface is viewed
as an innitesimally thin propagating surface. The key idea is to track the position of the ame front
using a eld variable G. The ame surface is associated to a given iso-level G = G . The G-eld does
not have to follow the gradients of the progress variable (e.g. temperature) and can be smoothed out to
be resolved on a LES grid. The challenge is to propose a model for the subgrid scale turbulent ame
163
speed, entering the ltered G-equation. According to Colin et al. [7] the G-equation approach is, up
to now, the most advanced technique for LES of premixed combustion. There also further references
can be found.
Filtering of the equation for progress variable Another approach, developed in Ref. [3] starts
from the balance equation for the progress variable. The ltered equation is solved on a mesh, but the
mesh size is chosen ner than the lter width. In this way the computational eciency of LES can be
combined with resolution of the gradients in the resolved progress variable. A ame surface density
formulation is proposed to describe the subgrid contribution of molecular diusive and reaction rate
processes. The quality of this closure has been investigated by comparing with data from a DNS.
Flame wrinkling density function The approach presented in Ref. [43] uses the laminar amelet
approach with conditional ltering. By conditioning the continuity equation on the unburned gas
before ltering, a transport equation for the resolved part of the unburned gas mass fraction is obtained.
From the transport equation for the subgrid ame area density and the resolved unburned gas volume
fraction, a transport equation for the subgrid ame wrinkling is derived. The subgrid ame wrinkling
can be regarded as the turbulent to laminar ame speed ratio.
9.3.3
Partially-premixed combustion
As outlined in this section the FGM approach is able to treat partially premixed combustion. However
in the past Domingo, Vervisch and Bray [8] have also developed LES models for partially premixed
ames in nonpremixed turbulent combustion. (I.e. when fuel and oxidizer enter the ow domain
separated and mix before burning). Combustion in the premixed mode is assumed to occur in the
thin amelet burning regime. Subgrid closure schemes for fully premixed and non-premixed combustion are combined. A criterion using the resolved unsteady fuel and oxidiser elds is introduced
to discriminate between the various degrees of partial premixing and to estimate the proportions of
premixed and non-premixed combustion at subgrid level.
9.4
Preferential diusion eects
In traditional models of premixed turbulent combustion preferential diusion is not taken into account.
However the eect can be quite large. The fact that all species diuse at a dierent rate which is
basically dierent from the diusion of heat and in combination with stretch eects makes it quite
complicated. Globally speaking there is of course the hydrodynamic instability which does not depend
on Lewis numbers. Besides that there is the important eect of thermo-diusive instabilities which is
a consequence of the dierence of diusion of the fuel compared to the heat (see Sivashinsky, [32]).
Because of the physical character this can be described by two variables, one for heat and one for
the fuel. The latter can be the progress variable. Such an approach is given in Vreman et al. [42].
It appears that the eect can increase the turbulent burning rate signicantly. In a more recent paper
we try to take more eects into account by looking at the basic changes that may occur in a ame
structure. Then a description in terms of enthalpy and element mass fraction becomes suitable (de
Swart et al. [36].
To go into a little more detail a rst analysis is presented on the behaviour of one-dimensional premixed ames including preferential diusion eects. First theory is presented, supplied by results of
detailed calculations of lean methanehydrogen mixtures. Both are based on the amelet equations as
164
presented in de Goey and ten Thije Boonkkamp, [15]. Flamelet equations are the governing equations
projected on a coordinate system attached to the ame. Therefore the one-dimensional coordinate is
dened as the trajectory normal to the gradients in a premixed ame. In this section we will consider adiabatic one-dimensional ames with dierent fuels and stretch rates. For diusion transport
modeling the constant Lewis number approach is used. In this approach the diusivity is taken to be
constant and expressed in terms of the conductivity, but for each species it has a dierent value. This
is consistent with our modeling work that we describe later in the paper.
Hydrogen addition will inuence the sensitivity of the mass burning rate to ame stretch. Let
us recall Eq. (49) of [15], where the stretched mass burning rate, m = sL , is written at the burnt
side. This mass burning rate, originally dened at the burnt side, can be rewritten at the inner layer
(subscript il) as
KaY,il
mil
=1
+ hil 0 ln m0 +
il
0
LeY
mil
hil
Ne 1
Z j,il
j=1
ln m0 ,
il
0
Z j,il
(9.27)
ignoring higher order terms. Here a superscript 0 indicates unstretched conditions, h is the enthalpy
and Z j is the element mass fraction of element j, for Ne elements. A ame progress is introduced with
Y, being a combination of chemical species such that it is monotonically increasing. It is normalized,
being 0 at the unburned side and 1 at burnt conditions. This solution is based on an analysis of the
amelet equations, Eqs. (20) of [15], with the Karlovitz integral dened as
KaY,il =
LeY
m0
il
sb
K[Y H(s sil )]ds ,
(9.28)
su
with H(s sil ) being the Heaviside function equal to zero when the distance along the ame s < sil
1
and one otherwise. The mass based ame stretch rate is K = M dM for an element of mass M, moving
dt
with the ame, see [14]. The preferential diusion eect, indicated by the two terms on the right hand
side of Eq. (9.27) is related to changes in enthalpy, hil and element mass fractionsZ j,il and depends
linearly on ame stretch, being zero at zero stretch. The preferential diusion terms can be written as
linear functions of KaY,il /LeY ,
hil
ln m0 +
il
0
hil
Ne 1
Z j,il
j=1
ln m0 = C KaY,il /LeY ,
il
0
Z j,il
(9.29)
which yields
KaY,il
KaY,il
mil
= 1 (1 C)
=1M
,
0
LeY
LeY
mil
(9.30)
and denes the Markstein number M = 1 C. The Markstein number describes the sensitivity of the
mass burning rate to ame stretch. Since higher order terms are not taken into account when dening
the Markstein number, the slope around K = 0 should be taken as the Markstein number. However,
since sometimes it is dicult to take the slope around K = 0 and approximately linear proles are
observed, in this study the Markstein number is taken as the slope of a complete set of data-points that
is considered. The Markstein number is equal to one, when there is no preferential diusion (Lei = 1).
Furthermore it must be recognized that a small value of the Karlovitz integral does not necessarily
indicate small stretch rates. Even in the thin reaction zones regime stretch might heavily uctuate
through the preheat and reaction zone (see e.g. [24]). However, positive and negative values of the
ame stretch easily cancel and result in a small Karlovitz integral while local ame stretch rates can
165
m/m0
1.5
0.5
0
1
0.5
0
Ka/Le
0.5
Figure 9.5: Inner layer mass burning rate as a function of Karlovitz integral for dierent methane/hydrogen
mixtures using detailed chemistry; full line XH2 = 0.0, dashed XH2 = 0.2, dash-dotted XH2 = 0.4.
be large. For weak stretch (and small variance through the ame), it is justied to neglect the higher
order terms and the eect of preferential diusion is captured by the Markstein number. However,
when stretch rates become larger, higher order Ka terms may become important and it is not certain
wether the relation as given in Eq. (9.30) yields unique Markstein numbers. This will be investigated
in later sections of this paper.
To study the eects of preferential diusion in one-dimensional ames a series of ames is simulated by using the one-dimensional code CHEM1D [35, 5], with detailed chemistry. Three dierent
methane/hydrogen fuel mixtures are considered, where the mole fraction of hydrogen XH2 is 0.0, 0.2
and 0.4, respectively. The stoichiometry is taken to be = 0.7 and all ames are calculated with
the GRI3.0 mechanism, [33]. The properties of these ames are shown in Table 9.1. It is shown that
when the hydrogen content in the fuel increases, the mass burning rate and the adiabatic temperature increase while the ame thickness decreases. The increase in mass burning rate is caused by an
increase in ame temperature and an increased supply of active radicals, as shown in e.g. [44, 11].
mixture A
mixture B
mixture C
XH2
0.0
0.2
0.4
XCH4
1.0
0.8
0.6
m0 [kg/m2 s]
0.218
0.250
0.300
T ad [K]
1844
1855
1871
0 [m]
f
6.77 104
6.12 104
5.37 104
Table 9.1: Properties of unstretched laminar ames with dierent fuel mixture, = 0.7.
To nd the Markstein numbers of the mixtures presented in Table 9.1, ame stretch and curvature
166
are applied in such a way that the mass burning rate is constant through the ame. The Markstein
number can then be found by plotting the scaled inner layer mass burning rate mil /m0 versus the
il
scaled Karlovitz integral KaY,il /LeY . This is shown in Fig. 9.5. One can observe that the absolute
slopes become smaller when XH2 in the fuel increases. The global slopes for the three cases are 1.01,
0.77 and 0.40, respectively. Taking local slopes at KaY,il /LeY = 0, yields 1.02, 0.75 and 0.35,
respectively. This shows that for weak stretch the slope around KaY,il /LeY = 0 in this ame stretch
range is close to the slope of the entire data-set. It becomes clear that the pure methane case shows
almost no preferential diusion eect but an increasing hydrogen content in the fuel mixture shows a
larger preferential diusion eect.
Concluding, it was shown that preferential diusion eects are large for hydrogen addition. This
means that in a computational approach, preferential diusion eects have to be accounted for. How
this is done in the framework of FGM, is explained in de Swart et al. [36].
References
[1] R.J.M. Bastiaans, J.A. van Oijen, S.M. Martin, L.P.H. de Goey, H. Pitsch DNS of lean premixed
turbulent spherical ames with a amelet generated manifold. CTR Annual Research Briefs,
257268, 2004.
[2] Bilger, R.W., Pope S.B., Bray, K.N.C., Driscoll, J.F.: Paradigms in turbulent combustion research. Proc. Comb. Inst. 30, 21-42 (2005).
[3] M. Boger, D. Veynante, H. Boughanem, and A. Trouv . Direct numerical simulation analysis of
e
ame surface density concept for large-eddy simulation of turbulent premixed combustion. In
Proc. Combust. Inst., 27, 1998.
[4] Bradley, D., Kwa, L.K., Lau, A.K.C., Missaghi, M., Chin, S.B.: Laminar amelet modeling of
recirculating premixed methane and propane-air combustion. Comb. Flame 71, 109-122 (1988).
[5] CHEM1D. A one-dimensional laminar ame code. Eindhoven University of Technology.
www.combustion.tue.nl/chem1d
[6] Clark, R.A., Ferziger, J.H., Reynolds, W.C.: Evaluation of subgrid-scale models using an accurately simulated turbulent ow. J. Fluid Mech. 91, 1-16 (1979).
[7] O. Colin, F. Ducros, D. Veynante, and T. Poinsot. A thickened ame model for large-eddy
simulations of turbulent premixed combustion. Phys. Fluids, 12:18431863, 2000.
[8] Pascale Domingo, Luc Vervisch and Ken Bray. Partially premixed amelets in LES of nonpremixed turbulent combustion. Combust. Theory Modelling, 6:519-551, 2002
[9] Driscoll, J.F.: Turbulent premixed combustion: Flamelet structure and its eect on turbulent
burning velocities. Progr. Energy Comb. Sci. 34, 91134 (2008).
[10] R.L.G.M. Eggels. Modelling of NOx formation of a premixed DLE gas turbine combustor. Proc.
TURBOEXPO 2001, ASME 2001-GT-0069, 2001.
[11] Gauducheau, J.L., Denet, B. and Searby, G. A numerical study of lean CH4/H2/air premixed
ames at high pressure. Combust. Sci. Technol., 137:8199, 1998.
167
[12] Germano, M., Piomelli, U., Moin, P., Cabot, W.: A dynamic subgrid-scale model. Phys. Fluids
A 3, 1760-1765 (1991).
[13] O. Gicquel, N. Darabiha, and D. Thevenin. Laminar premixed hydrogen/air counterow ame
simulation using ame prolongation of ILDM with dierential diusion. Proc. Comb. Inst., 28,
19011908, 2000.
[14] Goey, L.P.H de and Thije Boonkkamp, J.H.M. ten. A mass-based denition of ame stretch for
ames of nite thickness. Combust. Sci. Technol., 122:399405, 1997.
[15] de Goey, L.P.H. and ten Thije Boonkkamp, J.H.M. A amelet description of premixed laminar
ames and the relation with ame stretch. Combust. Flame, 119:253271, 1999.
[16] L.P.H. de Goey, T. Plessing, R.T.E. Hermanns, N. Peters. Analysis of the ame thickness of
turbulent amelets in the thin reaction zones regime. Proc. Combust. Inst. 30, 859866, 2005.
[17] E.R. Hawkes, R. Sankaran, J.C. Sutherland, J.H. Chen. Direct numerical simulation of turbulent
combustion: fundamental insights towards predictive models. J. Phys: Conference Series, 16,
6579, 2005.
[18] Leonard, A.: Energy cascade in large-eddy simulations of turbulent uid ows. Adv. Geophys.,
18, 237 (1974).
[19] Nicoud, F., Ducros, F.: Subgrid-scale stress modelling based on the square of the velocity gradient tensor. Flow, Turbulence and Combustion 62, 183-200 (1999).
[20] J.A.van Oijen. Flamelet-Generated Manifolds: Development and Application to Premixed Laminar Flames. PhD. Thesis, Technische Universiteit Eindhoven, 2002.
[21] J.A. van Oijen, L.P.H. de Goey. Modelling of premixed counterow ames using the ameletgenerated manifold method. Combust. Theory Modelling, 6(3), 463478, 2002.
[22] Van Oijen, J.A.: Flamelet-generated manifolds: development and application to premixed laminar ames. PhD. Thesis, TU Eindhoven (2002).
[23] van Oijen, J.A., Groot, G.R.A., Bastiaans, R.J.M. and de Goey, L.P.H. A amelet analysis of the
burning velocity of premixed turbulent expanding ames. Proc. Comb. Inst., 20:657664, 2005.
[24] van Oijen, J.A., Bastiaans, R.J.M., Groot, G.R.A. and de Goey, L.P.H. Direct numerical simulations of premixed turbulent ames with reduced chemistry: Validation and amelet analysis.
Flow, Turbulence and Combustion, 75:6784, 2005.
[25] N. Peters Turbulent Combustion. Cambridge University Press, 2000.
[26] Pitsch, H., Steiner, H.: Large-eddy simulation of a turbulent piloted methane/air diusion ame
(Sandia ame D). Phys. Fluids 12, 2541-2554 (2000).
[27] Thierry Poinsot and Denis Veynante Theoretical and Numerical Combustion. Second Edition,
Edwards Inc, 2005.
[28] S.B. Pope. Computations of turbulent combustion: progress and challenges. In Proc. Combust.
Inst., 23, 591612, 1990.
168
[29] Stephen B. Pope. Turbulent Flows. Cambridge University Press, 2000.

[30] Ramaekers, W.J.S.: The application of amelet generated manifolds in modelling of turbulent
partially-premixed ames. Master thesis, TU Eindhoven (2005).
[31] R. Sankaran, E.R. Hawkes, J.H. Chen, T. Lu, C.K. Law. Structure of a spatially developing
turbulent lean methane-air Bunsen ame. Proc. Comb. Inst. 31, 12911298, 2007.
[32] Sivashinsky, G.I., Instabilities, pattern formation, and turbulence in ames. Annu. Rev. Fluid
Mech., 15, 179199, (1983).
[33] Smith, G.P., Golden, D.M., Frenklach, M., Moriarty, N.W., Eiteneer, B., Goldenberg, M.,
Bowman, C.T., Hanson, R.K., Song, S. Gardiner Jr., W.C., Lissianski, V.V. and Qin, Z.,
http://www.me.berkeley.edu/gri mech/.
[34] Smooke, M.D., Giovangigli, V.: Formulation of the premixed and nonpremixed test problems. In
Reduced kinetic mechanisms and asymptotic approximations for methane-air ames, ed. M.D.
Smooke, Springer Verlag Berlin, 1-28 (1991).
[35] Somers, L.M.T. The simulation of Flat Flames with Detailed and Reduced Chemical Models.
PhD thesis, Technische Universiteit Eindhoven, Eindhoven, The Netherlands, 1994.
[36] de Swart, J.A.M., Bastiaans, R.J.M., van Oijen, J.A., de Goey, L.P.H. and Cant, R.S. Inclusion of
Preferential Diusion in Simulations of Premixed Combustion of Hydrogen/Methane Mixtures
with Flamelet Generated Manifolds. Flow Turbul. Combust. In press. (2010)
[37] Luc Vervisch and Denis Veynante. Turbulent combustion modeling. In Lecture notes, Von
Karman Institute, 1999. 177 pages.
[38] Vreman, A.W.: An eddy-viscosity model for turbulent shear-ow: algebraic theory and applications. Phys. Fluids 16, 3670-3681 (2004).
[39] Vreman, A.W., Bastiaans, R.J.M., Geurts, B.J.: Accuracy of large-eddy simulation of premixed
turbulent combustion, Flow, Turb. Combust. 82(2), 233-248, (2009).
[40] A.W. Vreman, B.A. Albrecht, J.A. van Oijen, L.P.H. de Goey, R.J.M. Bastiaans, Premixed and
non-premixed generated manifolds in large-eddy simulation of Sandia ame D and F, Comb. &
Flame, 153(3), 394-416, (2008)
[41] Vreman, A.W., van Oijen, J.A., de Goey, L.P.H., Bastiaans, R.J.M., Subgrid scale modeling in
large-eddy simulation of turbulent combustion using premixed amelet chemistry. Flow, Turb.&
Comb. 82, 511-535 (2009).
[42] Vreman, A.W., van Oijen, J.A., de Goey, L.P.H. and Bastiaans, R.J.M. Direct numerical simulation of hydrogen addition in turbulent premixed bunsen ames using amelet generated manifold
reduction. Int. J. Hydr. En., 34:27782788, 2009.
[43] H.G. Weller, G. Tabor, A.D. Gosman, and C. Fureby. Application of a ame-wrinkling LES
combustion model to a turbulent mixing layer. In Proc. Combust. Inst., 27, 1998.
[44] Yu, G., Law, C. K. and Wu, C. K. Laminar ame speeds of hydrocarbon + air mixtures with
hydrogen addition. Combust. Flame, 63:339347, 1986.
Chapter 10
Probability Density Function (PDF) methods

D.J.E.M. Roekaerts
In this lecture the joint velocity-scalar PDF approach to turbulent combustion modeling is described.
This approach was mainly developed by S.B. Pope [34] and includes from the start the complete onepoint joint statistics of velocity and scalars. This is conceptually appealing because it delivers in one
framework closure models for Reynolds stresses, turbulent scalar uxes and chemical source terms.
The joint velocity-scalar PDF approach is one of a set of possible PDF modeling approaches for
turbulent reaction ow (see the lecture on introduction to turbulent combustion modeling). Theory,
developments and applications of PDF methods over the last twenty ve year have recently been
reviewed by Haworth [14]. The PhD Theses of H. Wouters [42] and B. Naud [25] can be consulted
for further details on the specic variant of PDF methods presented here. Applications have been made
to the simulation of piloted jet diusion ames [30, 38, 28], blu body stabilized ames [40, 26, 27].
And the method has also been extended to simulation of evaporating and combusting sprays [25, 2].
10.1
PDF transport equation
From the conservation equations, the transport equation can be derived [34] for the joint velocityscalar PDF f U (V, ; x, t). By integrating this quantity over a range of values of V and the probability that U and take values in these ranges is obtained. It is also useful to consider the joint
velocity-scalar mass density function (MDF) dened as F U (V, , x; t) = () f U (V, ; x, t). The
transport equation for the MDF reads:
F +
Vi F
t
xi
Ti j
1
p
F
Vi ()
xi
x j
T i j
p
1
U = V, = F
Vi ()
xi
x j
1 Ji
+ S () F
() xi
J
i U = V, = F
()
xi
(10.1)
in which the rst two terms on the right hand side describe the evolution in velocity space and the
last two terms describe the evolution in scalar space. (The notation A| B denotes the conditional
expectation value of A upon condition B). The terms that contain the mean pressure gradient and the
mean viscous stress tensor can be expressed in terms of the mean velocity eld and thus occur in closed
169
170
form. Also the reaction rate occurs in closed form. The unclosed terms describe the mean eects of
the uctuating pressure gradient and the uctuating viscous stress tensor, conditional on the values
of velocity and scalars. Assuming high Reynolds number, eects of mean shear T i j are usually
neglected. The terms containing the uctuating pressure gradient and uctuating viscous stress tensor
contains unclosed conditional averages and cannot be expressed in terms of the one-point distribution
of U and . The terms that describe the evolution in scalar space contain the eects of the mean
molecular diusion ux and of reaction. The reaction rate can be expressed as a function of the scalar
variables and therefore it is contained in closed form. The term containing the conditional mean
eects of the uctuating diusion ux describes molecular mixing or micro-mixing and is unclosed.
Since mixing by diusion is an essential process bringing reactants and/or heat together the modelling
of this term is important. It is directly related to the conditional scalar dissipation rate and connections
with other approaches to turbulent combustion modeling can be formulated on the basis of this.
Starting from the transport equation for the joint velocity-scalar mass density function it is straightforward to derive the equation for the joint velocity-scalar Favre PDF f U . Both formulations are
equivalent. The Eulerian transport equations for the moments of f U can be derived by multiplying
each term by a function of Vi and (e.g., Vi V j , Vi V j Vk , , or Vi ) and integrating over the
phase space (V, ). In this way also a transport equation for the Reynolds stresses can be derived and
it is a good question to ask which Eulerian second-moment closure model is implied by a choice of
closure of the PDF transport equation, or the other way round, to ask whether a proposed second moment closure model has a Lagrangian counterpart. Such questions are discussed in detail by Pope [35]
and Wouters [41]. Similar question also arise concerning the modeling of the turbulent scalar ux.
The modelled velocity-scalar PDF equation implies an equation for the turbulent scalar ux. Further
developments of this topic can be found in Ref. [29].
10.2
Monte Carlo solution method
The PDF transport equation (equation 10.1) is a partial dierential equation in many dimensions. In
the case of d spatial dimensions and n independent scalars, solving the stationary equation by means of
a nite-dierence technique would require a suciently accurate discretization in 2d + n dimensions.
Alternatively, in a Monte Carlo solution method, the 2d + n dimensional mass density function is
represented by a large ensemble of N notional uid particles. Here the term notional uid particles
is used to distinguish from real physical uid elements. The particle properties evolve according to
particle models such that the evolution of the statistics of the particle ensemble corresponds to the
modeled PDF evolution. It is not required that single notional particle trajectories represent real uid
element trajectories. The Monte Carlo method is computationally more ecient already for the case
d = 2, n = 1. Furthermore it oers additional opportunities for model development. The basics
of the Monte Carlo method and several of the numerical algorithms involved are described in detail
by Pope [34], Haworth and El Tahry [13], Correa and Pope [3], and Wouters [41]. This involves
a.o. the method of fractional steps, the estimation of mean elds and mean eld gradients, particle
boundary conditions and the integration of particle evolution equations. Here we consider systems
whose one-point statistics depend on two spatial coordinates only (d = 2). All Eulerian quantities,
like mean velocity and scalar elds and the mass density function, are dened on a two-dimensional
grid. The Eulerian joint velocity-scalar mass density function is represented by a spatially equally
distributed ensemble of Lagrangian notional uid particles with properties like position, velocity
and composition.
171
Each of the particles represents an amount of mass m and the particles are distributed such that
the distribution of mass in space corresponds to the actual density eld. An estimate of the Eulerian
Favre joint velocity-scalar PDF or, equivalently, the joint velocity-scalar mass density function in a
cell is given by the velocities and scalar properties of the particles present in that cell. Note that the
mass density function is dened on a grid but that the particles are not. The particles each have their
own position and velocity and are not bounded to a grid. The estimation of the Eulerian PDF from the
ensemble of Lagrangian particles is illustrated in gure 10.1.
The requirement that the particle ensemble gives a physically correct estimate of the mass density
function, is expressed mathematically by the requirement that the sum of particle masses divided by
the cell volume, is equal to the mean the density eld. In axisymmetrical congurations, the two
spatial coordinates are axial and radial position. Then cells near the axis represent a small volume in
three dimensions whereas cells far from the axis correspond to a larger three-dimensional volume. If
all particles are to represent the same amount of mass m, the particle number in a cell must depend
on the local density and on the three-dimensional cell volume. Consequently, in cells with a small
volume or with low density the number of particles is lower and the statistical accuracy in these cells
is lower. As a result, the statistical accuracy varies throughout the domain which is undesirable. To
remedy this eect an ensemble with variable particle weights w(i) m is used allowing the use of a
nearly constant number of particles per cell. Then the average in Equation 10.5 has to be replaced by
a weighted average.
Figure 10.1: Illustration of the Eulerian PDF estimate fN (; x) from an ensemble of Lagrangian particles (i)
in a cell at position x.
The particle properties evolve according to a modeled Lagrangian system of the following general
form:
dxi = Ui dt
dUi =
d =
Kiu dt + bi j dW j
K dt + b dW
(10.2)
(10.3)
(10.4)
The drift vectors K u and K for respectively the velocity and scalars describe a deterministic
change, the terms containing the increment of the Wiener process dW describe a stochastic change,
in the form of a diusion process (see below). All Wiener increments in 10.2 are independent of each
other. The specic expressions for the drift and diusion terms for velocities and scalars dene the
model. Examples of particle velocity and scalar evolutions are given below.
The coecients of the particle models contain two dierent kind of mean quantities. First there
are the Eulerian mean quantities, like Ui and . These can be estimated from the particle properties
172
EULERIAN SUBMODEL
LAGRANGIAN SUBMODEL
Lagrangian Algorithm
p
k
xi
Finite Volume Method
solve for
particle properties:
x(i) , U(i) , (i)
solve for:
mean elds
U, p, ui u j , k,
(i)
(i)
(i) T (i)
t
?
Chemistry Scheme
Figure 10.2: Sketch of the hybrid Monte Carlo method. Left side: Lagrangian submodel with chemistry solver.
Right side: Eulerian submodel that solves for mean ow eld.
e.g. by the following expression:

1
=
Nc
Nc
(i)
(10.5)
i=1
in which the summation is over all particles present in a cell. (In practice also smoothing algorithms
are used) Second, the modeled terms contain a characteristic mean turbulent frequency that determines
the rate at which the modeled processes take place. This frequency cannot be estimated from the
particle properties in the joint velocity-scalar description and it has to be supplied externally.
The mean pressure eld can be obtained from the velocity and density elds and, in principle, it
is closed. Direct calculation of the mean pressure eld would require a pressure solver that uses the
particle velocity and mass distributions as input. Here a hybrid Monte Carlo method is employed in
which both the pressure and turbulent frequency elds are solved by an external Eulerian nite-volume
method according to the method rst proposed by Correa and Pope [3].
10.3
Hybrid ow eld model
The hybrid Monte Carlo method consists of a Eulerian nite-volume submodel and a Lagrangian
Monte Carlo submodel. The hybrid algorithm is illustrated in gure 10.2. The Eulerian model solves
for the mean velocity, pressure, turbulent kinetic energy and turbulent dissipation. Given these mean
elds, the Lagrangian Monte Carlo method solves for the joint velocity-scalar mass density function.
Eects of thermo-chemistry are solved in the Lagrangian algorithm. To enhance the computational
173
eciency of the calculation of the chemical eects, dierent techniques can be combined. The dimension of the phase space can be reduced by using reduced chemical mechanisms and lookup tables
and ecient storage and retrieval algorithms can diminish the cost of the evaluation of the source term
of a given mechanism.
Mean thermo-chemical properties that inuence the ow eld, like density and molecular viscosity elds, are coupled back to the nite-volume method which solves for the new turbulent ow eld.
Subsequent submodel calculations are performed until a stable solution is reached. In this algorithm,
conventional modeling of the Eulerian ow eld equations, in particular the Reynolds-stress equations, is needed also, and closer inspection is needed of the relation between closure in the Lagrangian
and the Eulerian framework.
10.3.1
Eulerian submodel
The mean momentum equation is closed using a dierential Reynolds stress model. Assuming
high Reynolds number, viscous terms are neglected except for the viscous dissipation term i j . The
Reynolds stress equations for variable density ows then read:
D
u u
Dt i j
U j
Ui
= ui uk
u j uk
xk
xk
p
p
p
2
ui
x + u j x 3 i j uk x
j
i
k
ij
ui u j uk + i j uk p
xk
3
p
p
ui
uj
x j
xi
u
2
+ i j p k
3
xk
(10.6)
Here the Reynolds average is denoted by an overbar and the combined use of an overbar and
tilde, as in x, is used to denote the Favre average of a long expression. The terms on the RHS are:
the production by mean shear Pi j , the pressure-strain correlation i j , the viscous dissipation i j , the
turbulent ux Ti j , and two terms which are zero in constant density ows containing a mean pressure
gradient, and the trace of the uctuating strain tensor. The production terms are in closed form whereas
the uctuating pressure, dissipation, turbulent ux, and uctuating density terms have to be modeled.
The viscous dissipation i j is modeled by assuming local isotropy at the smallest scales where
viscous dissipation takes place. The dissipation model then reads:
ij
For the dissipation of turbulent kinetic energy
2
i j
3
(10.7)
a standard modeled equation is solved. The triple
correlation terms ui u j uk present in the ux terms can be modelled by a generalized gradient diusion
model that reads:
ui u j
k
ui u j uk = C s uk ul
(10.8)
xl
174
where the constant C s has a value around 0.25. Modeling of the uctuating density terms can be
found in the papers by Jones [19, 20]. The nal unclosed term is the pressure-strain redistribution
term. This term does not produce or destroy turbulent kinetic energy but only redistributes energy
over the components of the stress tensor. Several closure models proposed for this term are reviewed
in [?].
10.3.2
Langevin model for the modeling of the particle velocity
The unclosed terms in the joint velocity-scalar mass density function equation that describe the evolution in velocity space read:
T i j
U = V, = f
(10.9)
Vi
xi
x j
here written in terms of the Favre joint PDF instead of the mass density function. To illustrate the
closure procedure, we rst consider the Simplied Langevin Model (SLM) [12]. A Langevin model
is a stochastic model for uid particle velocities. The modeled particle velocity equation consists of a
linear drift towards the local Favre mean and an added isotropic diusion term, and reads:
dUi =
1 3
+ C0 (Ui Ui )dt + (C0 )1/2 dWi
2 4
(10.10)
in which = /k is the mean turbulent frequency, Ui is a stochastic velocity realization, C0 a constant

with a physical meaning explained below , and dWi a stochastic Wiener increment. The Wiener
process is a stochastic process with zero mean and covariance:
dWi dW j = dti j
(10.11)
The Wiener or diusion process is continuous but not dierentiable and has the properties of a Markov
process. This means that the stochastic velocity increments only depend on the present state of the
uid element and not on its history. Using Ito calculus the Langevin model then corresponds to an
evolution of the velocity PDF given by Risken [37], and Gardiner [10]:
1 3
2 f U
fU
1
=
+ C0
(Vi Ui ) f U + C0
t
2 4
Vi
2
Vi2
(10.12)
The basic properties of the SLM are that it yields the correct evolution of turbulent kinetic energy and
that it yields isotropization of the velocity distribution. To describe complex ow elds an extended
model is needed: the Generalized Langevin model (GLM) which is of the form
dUi =
Ti j
1 p
xi
x j
dt
+Gi j (U U j )dt + (C0 )1/2 dWi

j
(10.13)
in which C0 is a positive constant, W(t) an isotropic Wiener process, the dissipation of turbulent
kinetic energy and Gi j is a second-order tensor which is modeled as a function of local mean quantities.
A general form of Gi j in terms of the local mean velocity gradients, Reynolds stresses and dissipation
was derived by Haworth and Pope [12].
The reason for using the Langevin model to model turbulent velocities is found in the consistency
of the model with Kolmogorovs hypothesis of turbulence [23]. For high Reynolds number there is a
175
separation between the large energy containing scales and the small scales at which viscous dissipation
takes place. The cascade of energy from the large to the small scales occurs in the inertial subrange
where the energy dissipation rate is the only relevant parameter. The eects modeled by the tensor
Gi j are characterized by the time-scales of mean deformation | U /x|1 and energy dissipation
T = k/ . The Wiener process models eects that take place on a shorter time-scale. The specic form
of the diusion coecient is determined by the modeled Lagrangian structure function which, for the
Langevin model, reads:
Ui U j
(Ui (t + s) Ui (t))(U j (t + s) U j (t))
= C0 si j
(10.14)
with Ui the velocity increment of a Lagrangian uid element. This is consistent with Kolmogorovs
inertial range scaling with C0 being a universal constant. A value C0 = 2.1 was determined experimentally from diusion measurements in grid turbulence by Anand and Pope [1]. For a detailed
treatment of dynamical properties of the Langevin model and consistency with Kolmogorovs theory,
see [11, 12, 36].
An important feature of the Lagrangian models is that these describe the evolution of individual
uid element velocities. This means that, for nite model coecients, the Langevin model always
describes physically realizable distribution of velocities (e.g. positive normal stresses and turbulent
kinetic energy). Consequently, the Langevin model is always realizable by construction as are all
moment equations derived from the Langevin model.
Because the Langevin model describes the evolution of the full joint velocity distribution it also
provides a modeled evolution equation for the second moments of velocity. In particular, in the case
of an isotropic dissipation tensor ( i j = 2 i j ) the exact expression for the pressure strain correlation
3
in the Reynolds stress equation (10.6)
j
i
1 p
p
= i
x + u j x
j
i
(10.15)
is represented by
j = Gik u j uk + G jk ui uk +
i
2
+ C0
3
i j
(10.16)
In the approach followed here, following the lines of Pope [35], the model parameters in the
expression for Gi j are specied such that the modeled pressure-strain correlation equals that of one of
several well-known second-moment closures.
A basic second-moment closure model for j is the combination of a Linear return-to-isotropy
i
or Rotta term and a linear isotropization-of- production (IP) term. The model (with xed model
constants C1 = 1.8 and C2 = 0.6) is referred to as the Isotropization-of-Production Model (IPM) and
the corresponding GLM is called the Lagrangian IP model (LIPM). The Simplied Langevin model
(SLM) corresponds to the Rotta model (only linear return-to-isotropy C1 = 4.15, C2 = 0). For the
(L)IPM and the SLM/Rotta model, hybrid Monte Carlo calculations were performed in a blu-body
stabilized diusion ame using both the Eulerian second-moment model and the Lagrangian Langevin
models [40]. Some results of these calculations are presented below.
Since the full joint probability of velocities and scalars is known the Lagrangian models also provide modeled evolution equations for the third-moments of velocity and the turbulent scalar uxes.
The third-moment equations are governed by the Langevin model for velocities whereas the combination of Langevin model and micro-mixing model provides a model for the scalar uxes.
176
10.3.3

Consistency
When solving the modeled velocity PDF transport equation with the hybrid Monte Carlo method some
caution is needed. In the hybrid method the turbulent ow eld is represented in two ways: on the one
hand by the Eulerian mean velocity and turbulent kinetic energy elds of the nite-volume submodel,
and on the other hand by the velocity distribution of the Lagrangian Monte Carlo submodel. Although
a unique denition of the turbulence model is not guaranteed, both submodels use the same solution
for the mean velocity and turbulent kinetic energy elds. Turbulent transport of mean momentum and
production of turbulent kinetic energy are governed by the Reynolds stresses and consistency between
the two submodels on the level of mean velocities and kinetic energy inevitably requires a consistent
treatment of the second moments.
Several studies, using the hybrid method, have used the k- model in combination with the
SLM [4, ?]. In the Eulerian k- model, the Reynolds stresses are given by an algebraic relation and
the second-moment equations are not modeled explicitly. No equivalent model in terms of a Langevin
model exists. As a result, the k- model cannot be part of a consistent hybrid model. In the hybrid
method employed here, a consistent hybrid turbulence model is obtained by using the same secondmoment equation in both submodels with the Langevin model chosen equivalent to the nite-volume
turbulence model for the pressure-strain correlation. The third-moments are still treated inconsistently
in the two submodels. In the Lagrangian method the third-moments are exact whereas in the second
moment closure models these are given by a generalized gradient diusion model. Eects of this
inconsistency on the level of the third moments are assumed to be small (See [25] for a conrmation
of this assumption).
Fully consistent hybrid algorithms, i.e. eliminating the problem with the third moments just mentioned, have also been developed, see Refs. [24, 18] and references therein. In these methods Reynolds
stresses are calculated in the Monte Carlo submodel only.
10.4
10.4.1
Modeling of scalar micro-mixing

Introduction
The process of micro-mixing is a very important mechanism in turbulent reacting ows. On the
largest scales mixing of species takes place by turbulent convection but only after species are mixed
by molecular diusion on the smallest scales, chemical reaction can occur. For Damk hler number
o
around unity, turbulent mixing and reaction are strongly coupled whereas for high Damk hler number
o
the overall reaction rate is limited by the mixing process. In both cases, for a prediction of the mean
chemical source terms an accurate description of the scalar distribution is needed. Micro-mixing
models are designed to mimic several physical properties of molecular scalar mixing. These properties
stem either from the Navier-Stokes equations or are motivated by observations in measurements or
DNS of inert scalar mixing. Below we start from exact equations for scalar mixing and list several
basic properties of turbulent scalar mixing. Then the two most widely used micro-mixing models are
reviewed: Interaction by Exchange with the Mean and Coalescence/Dispersion.
10.4.2
Turbulent scalar mixing
Exact equations
The evolution of the Favre joint velocity-scalar PDF under the inuence of turbulent molecular scalar
mixing is given by equation 10.1. The unclosed term consists of the mean molecular diusion ux
177
conditional on the velocity and scalar values. The evolution of the PDF under the inuence of micromixing is given by:
Ji
1
U = V, = f U
(10.17)
f U =
t

xi
Assuming Fickian diusion and equal and constant diusivities, the term on the RHS can be rewritten
as:
2
2
D 2 f U D
U = V, = f U
(10.18)

xi xi
xi
The rst term describes the diusion in physical space and is usually neglected assuming high Reynolds
numbers. Using the same argument the eects of the mean diusion ux J are neglected also. This
j
term (see equation 10.1) contains only mean eld gradients and is contained in closed form. The
remaining unclosed term is the conditional scalar dissipation or cross-dissipation which is dened by:
,
=D

U = V, =
xi xi
(10.19)
The eects of this term on the scalar distribution can be summarized by a decay of scalar variance
combined with a relaxation of the scalar PDF to a Gaussian-like shape. Whichever model is used to
mimic these eects, all models require an additional time-scale that determines the scalar variance decay rate. The decay of scalar (co)-variance is given by the averaged, unconditional, scalar dissipation
or cross-dissipation which is given by:
d
= 2
dt
(10.20)
The closure of the variance decay rate also arises in modeling of the second-moment equations. For
single inert scalar mixing the scalar variance decay rate most often is modeled to be proportional to
the decay rate of turbulent kinetic energy where the constant of proportionality C takes a standard
value of 2.0.
Properties of turbulent scalar mixing
The rationale behind the use of a particular model for micro-mixing lies in the ability to mimic relevant
physical properties of the mixing process. Here several important properties of the mixing process are
reviewed: (1) the scalar mean is conserved, (2) correct variance decay rate, (3) boundedness of scalars,
(4) linearity and independence, (5) relaxation to a Gaussian for inert scalar mixing, and (6) localness
in composition space.
Constant scalar mean In the absence of a mean diusion ux, molecular scalar mixing does not
aect the mean scalar value. Conservation of the scalar mean is a strictly required property of mixing
models. Models that do not satisfy this property may, for example, destroy or produce a conserved
scalar like mixture fraction; violating mass and element balances.
Scalar variance decay Under the assumptions of high Reynolds numbers, the scalar variance equation shows that the decay of scalar variances by molecular diusion is given by the scalar dissipation
. The joint velocity-scalar PDF does not provide information on the decay rate and the decay rate
must be supplied externally. All micro-mixing models are designed to reproduce the available information on the decay rate.
178
Boundedness of scalar values An important property of scalar mixing is that mixing cannot cause
scalar values to occur below the minimal or above the maximal values. For example: in a region
with mixture fraction values between 0.5 and 1, molecular mixing cannot cause scalar values to occur
below 0.5 or above 1. In more dimensions, consider a bounded region which contains all scalar
values. During a mixing step, all scalar values must remain within the bounds of the previously
spanned region. Boundedness must be satised to prevent unphysical scalar values like negative mass
fractions or temperature.
Linearity and independence The scalar transport equations, for non-reacting scalars, are linear in
the scalar values. This implies that, for equal diusivities, any linear combination of non-reacting
scalar variables evolves according to the same set of equations. A good micromixing model should
respect this linearity principle. Related to this, the independence principle implies that, in mixing, the
evolution of the statistics of one scalar are not inuenced by the distribution of the other scalars [33].
The models reviewed here satisfy these two fundamental properties.
Relaxation to a Gaussian in homogeneous turbulence Turbulent scalar mixing does not only result
in a decay of scalar variance but the PDF shape relaxes to a bell-shaped distribution. For mixing of
temperature uctuations in grid-generated turbulence [17] and in DNS simulations of inert mixing
homogeneous isotropic turbulence [8] it has been observed that an initially binary distribution evolves
to a Gaussian. In computations of mixing in inhomogeneous ows, like jet diusion ames, it may
not be so relevant whether or not a model reproduces the exact Gaussian shape for the homogeneous
test case. But models that can reproduce the DNS data are favored.
A good measure of the evolution of the PDF shape is the evolution of the normalized central
moments of a distribution:
n
n =
(10.21)
(2 )n/2
with n = n . For a binary distribution 4 = 6 = 1 and for a standard Gaussian distribution 4 = 3
and 6 = 15. The relaxation of the scalar PDF from a binary distribution is measured by the evolution
of these moments in time and by the way in which the Gaussian values are approached.
Localness in scalar space The concept of localness in multidimensional scalar space was introduced by Subramaniam and Pope [39] (for a single scalar the concept already appears in Norris and
Pope [31]). The additional requirement introduced is that mixing of scalar particles is governed by its
close neighborhood in composition space. A motivation for it is that the mixing of uid particles in
a physical volume of uid, where scalar elds are continuous, is also local in scalar space. However,
it may not be necessary to satisfy this requirement in order to represent mixing of the scalar PDF
correctly. The mixing models reviewed here do not satisfy the localness condition.
10.4.3
Modeling of micro-mixing
In this section two most widely used micro-mixing models are described: Interaction by Exchange
with the Mean model and Coalescence/Dispersion or Curls Models. For a discussion of other models
we refer to the literature, e.g. S. Heinz [15] and R.O. Fox [9]. S. Subramanian and S.B. Pope have
introduced the Euclidean Minimal Spanning Tree model [39] (EMST). D.W. Meyer and P. Jenny have
introduced the Presumed Scalar Prole model [22]. The comparative study of the relative merits of
all models is beyond the scope of this lecture. H.A. Wouters [41] has compared IEM and C/D with
179
mapping closure and Binomial Langevin model. Merci et al. [21] have compared IEM, C/D and
EMST. Other comparative studies are mentioned in the review by Haworth [?].
Interaction by Exchange with the Mean model
The simplest micro-mixing model is the Interaction by Exchange with the Mean (IEM) or Linear Mean
Squares Estimate (LMSE) model [6, 7]. The IEM model is widely used because of its simplicity and
ability to mimic the most important eects of mixing. The model conserves the scalar mean and yields
the correct variance decay rate.
The decay of scalar variance is modeled by a deterministic relaxation of all scalar values to the
local mean.
1
d = ( )dt
(10.22)
2
in which is the modeled scalar variance decay rate. The IEM model corresponds to an evolution
of the scalar PDF given by:
f 1
(10.23)
=
( ) f
t
2
In the multi-scalar formulation given here, all scalars relax to the respective mean at the same decay
rate .
With respect to the physical properties of turbulent scalar mixing that were mentioned in the
previous section, the IEM model only mimics the rst four properties; conservation of the mean, a
correct variance decay rate, boundedness of scalars, and linearity and independence. The model does
not predict any relaxation of the shape of the scalar distribution. The predicted evolution of normalized
moments is dn /dt = 0 for all n.
Coalescence/Dispersion models
Coalescence/Dispersion (C/D) or particle-pair interaction models are most easily dened in terms of
a particle model where mixing takes place in randomly selected particle pairs. The original model
was proposed by Curl [5]. Improvements of Curls model were proposed by Janicka et al. [16], and
Pope [32], the extension of the model to the case with variable particle weights is given in Nooren
et al. [?], and Wouters [41]. Coalescence/Dispersion models are attractive because the models are
relatively simple to use. Furthermore, C/D models are able to mimic the physics of mixing in a better
way than the IEM model does and perform reasonable well for mixing of multiple reacting scalars.
The standard C/D model for scalar particles is dened as follows: consider an ensemble of N
particles with equal weights w(i) = 1 and scalar values (i) . For every mixing step, two randomly
chosen particles (p, q) mix with a certain probability Pmix . The particle pair mixes towards the pair
mean scalar value:
(p,q)
(p) (t + t) = (q) (t + t) = ,mean =
1
(p) (t) + (q) (t)
2
(10.24)
while the other particles remain unchanged. For a selected particle pair, all scalars take part in
mixing. For this case of equally weighted particles the decay of variance in a mixing step t is given
by:
Pmix
g (t + t) = g (t) 1
(10.25)
N
180
with g an arbitrary scalar variance. In the limit t 0 the correct decay of variance is obtained if
Pmix is chosen:
Pmix = N[1 exp( t)] Nt
(10.26)
The numerical implementation of this algorithm is straightforward.
The PDF shapes predicted by this model in case of mixing of a binary PDF are unrealistic. To
remedy the latter eect, Janicka et al. [16] propose a modication of the model in which a particle
pair only exchanges a fraction of the scalar values. This fraction is chosen randomly from a uniform
distribution. The modied C/D model yields a relaxation of the PDF shape to a bell-shaped function
but the normalized skewness and super-skewness do not approach the Gaussian values. To improve the
asymptotic behavior of the C/D model, Pope [32] proposes an extension with an age-biased selection
of particles. This results in a probability of selection of particles that is biased to particles with a
larger age, where age is dened as the time a particle has not taken part in mixing. Improvements as
proposed by Pope have been tested only for the test case of binary mixing. Although the modied
C/D model with A(a) = 1 overpredicts the tails of the PDF, with innite moments 4 and 6 , the
PDF is bell-shaped and the disagreement with a Gaussian PDF is noticeable only in the low tails of
the distribution. In all most publications in the literature, including [21], the modied C/D model by
Janicka et al. [16], extended for particles with unequal weights [?], is used.
Summarising, with respect to the physical properties of turbulent scalar mixing, the C/D model
conserves the scalar mean, yields the correct variance decay rate, and satises the requirements of
boundedness, linearity and independence. For homogeneous inert mixing the PDF evolves to a bellshaped function but does not approach an exact Gaussian.
10.5
Concluding remarks
In the exact transport equation for the one-point velocity-scalar PDF the mean chemical source term
appears in closed form. The solution of the PDF equation is most eciently done using a stochastic
particle method. This method is computationally more expensive than simpler closure models (laminar nite rate chemistry, Eddy Dissipation Concept) but give a much more precise representation of
the inuence of turbulent uctuations on the mean reaction rate.
Provided the chemistry is represented in sucient detail for a specic application, the accuracy of
the modeling of the micromixing term remains a factor limiting the accuracy of the modeling results.
References
[1] M.S. Anand and S.B. Pope. Diusion behind a line source in grid turbulence. In L.J.S. Bradbury
et al., editor, Turbulent Shear Flows, volume 4, pages 4661. Springer-Verlag, New York, 1985.
[2] N. Beishuizen, B. Naud and D. Roekaerts. Evaluation of a modied Reynolds stress model
for turbulent dispersed two-phase ows including two-way coupling. Flow, Turbulence and
Combustion, 79: 321341, 2007.
[3] S.M. Correa and S.B. Pope. Comparison of a Monte Carlo PDF/nite-volume mean ow model
with blu-body Raman data. In Proceedings of the Combustion Institute, 24, pages 279285,
1992.
181
[4] S.M. Correa, A. Gulati, and S.B. Pope. Raman measurements and joint PDF modeling of a
nonpremixed blu-body-stabilized methane ame. In Proceedings of the Combustion Institute,
25, pages 11671173, 1994.
[5] R.L. Curl. Dispersed phase mixing: I. Theory and eects in simple reactors. Journal of the
American Institute of Chemical Engineers, 9(2):175181, 1963.
[6] C. Dopazo. Relaxation of initial probability density functions in the turbulent convection of
scalar elds. The Physics of Fluids, 22(1):2030, 1979.
[7] C. Dopazo. Recent developments in PDF methods. In P.A. Libby and F.A. Williams, editors,
Turbulent reacting ows, pages 375474. Academic press, London, 1994.
[8] V. Eswaran and S.B. Pope. Direct numerical simulations of the turbulent mixing of a passive
scalar. The Physics of Fluids, 31(3):506520, 1988.
[9] R.O. Fox Computational models for turbulent reacting ows. Cambridge University Press,
Cambridge, UK, 2003
[10] C.W. Gardiner, Handbook of stochastic methods. Springer-Verlag, 2nd Edition, Berlin, 1990.
[11] S.S. Girimaji and S.B. Pope. A diusion model for velocity gradients in turbulence. Physics of
Fluids A, 2(2):242256, 1990.
[12] D.C. Haworth and S.B. Pope. A generalized Langevin model for turbulent ows. The Physics
of Fluids, 29:387405, 1986.
[13] D.C. Haworth and S.H. El Tahry. Probability density function approach for multidimensional
turbulent ow calculations with application to in-cylinder ows in reciprocating engines. AIAA
Journal, 29(2):208218, 1990.
[14] D.C. Haworth Progress in probability density function methods for turbulent reacting ows.
Progress in Energy and Combustion Science, 36:168259, 2010.
[15] S. Heinz Statistical mechanics of turbulent ows Springer-Verlag, Berlin, Heidelberg, New
York, Tokyo, 2003
[16] J. Janicka, W. Kolbe, and W. Kollmann. Closure of the transport equation for the probability
density function of turbulent scalar elds. Journal Non-Equilibrium Thermodynamics, 4:4766,
1979.
[17] Jayesh and Z. Warhaft. Probability distribution, conditional dissipation, and transport of passive
temperature uctuations in grid-generated turbulence. Physics of Fluids A, 4(10):22922307,
1992.
[18] P.Jenny, S.B. Pope, M. Muradoglu and D.A. Caughey. A hybrid algorithm for the joint pdf
equation of turbulent reactive ows. J. Comput. Phys., 166:218252, 2001.
[19] W.P. Jones. Models for turbulent ows with variable density and combustion. In W. Kollmann,
editor, Prediction methods for turbulent ows, pages 379421. Hemisphere, Washington DC,
1980.
182
[20] W.P. Jones. Turbulence modelling and numerical solution methods for variable density and
combusting ows. In P.A. Libby and F.A. Williams, editors, Turbulent reacting ows, pages
309374. Academic press, London, 1994.
[21] B. Merci, D. Roekaerts, B. Naud, and S.B. Pope. Comparative study of micromixing models in
transported scalar PDF simulations of turbulent non premixed blu body ames. Combustion
and Flame, 146:109130, 2006
[22] D.W. Meyer and P. Jenny. Micromixing models for turbulent ows. J. Comput. Phys., 228:1275
1293, 2008
[23] A.S. Monin and A.M. Yaglom. Statistical Fluid Mechanics. MIT Press, Cambridge, Massachusetts, 1971.
[24] M. Muradoglu, P. Jenny, S.B. Pope, and D.A. Caughey. A consistent hybrid nitevolume/particle method for the PDF equations of turbulent reactive ows. Journal of Computational Physics, 154:342371, 1999.
[25] B. Naud. PDF modeling of turbulent sprays and ames using a particle stochastic approach.
PhD thesis, Delft University of Technology, 2003.
[26] B. Naud, C. Jimenez, and D. Roekaerts. A consistent hybrid PDF method: implementation
details and application to the simulation of a blu-body stabilised ame. Progress in Computational Fluid Dynamics, 6:146157, 2006
[27] B. Naud, B. Merci, D. Roekaerts, D. Schmidt and U. Maas. Scalar PDF and velocity-scalar
PDF modelling of the blu-body stabilised ame HM1 using ILDM. Proc. of the 5th Int. Symp.
Turbulence, Heat and Mass Transfer, K. Hanjalic, Y. Nagano and S. Jakirlic (Eds.), Begell
House, ISBN 1-56700-229-3, 2006.
[28] B. Naud, C. Jimenez, B. Merci, and D. Roekaerts. Transported PDF calculations of the piloted jet
diusion ame Delft ame III with complex chemistry: study of the pilot ame model. In Dirk
Roekaerts, Pedro Coelho, Bendiks-Jan Boersma and Kilian Claramunt, (Editors), Proceedings
of the 2nd ECCOMAS Thematic Conference on Computational Combustion, 2007, Delft: TU
Delft. ISBN 978-90-811768-1-1
[29] B. Naud, D. Roekaerts and B. Merci. Generalised Langevin model in correspondence with a
chosen standard scalar-ux second-moment closure. Flow, Turbulence and Combustion, 2010,
DOI10.1007/s10494-010-9282-3
[30] P.A. Nooren, H.A. Wouters, T.W.J. Peeters, D. Roekaerts, U. Maas, and D. Schmidt. Monte
Carlo PDF modelling of a turbulent natural-gas diusion ame. Combustion Theory and Modelling, 1(1):7996, 1997.
[31] A.T. Norris and S.B. Pope. Turbulent mixing model based on ordered pairing. Combustion and
Flame, 83:2742, 1990.
[32] S.B. Pope. An improved turbulent mixing model. Combustion Science and Technology, 28:131
145, 1982.
[33] S.B. Pope. Consistent modeling of scalars in turbulent ows. The Physics of Fluids, 26:404408,
1983.
183
[34] S.B. Pope. PDF methods for turbulent reactive ows. Progress in Energy and Combustion
Science, 11:119192, 1985.
[35] S.B Pope. On the relationship between stochastic Lagrangian models of turbulence and secondmoment closures. The Physics of Fluids, 6(2):973985, 1994.
[36] S.B Pope. Lagrangian PDF methods for turbulent ows. Annual Review Fluid Mechanics,
26:2363, 1994.
[37] H. Risken. The Fokker-Planck equation. Springer-Verlag, Berlin, 1984.
[38] Dirk Roekaerts, Bart Merci and Bertrand Naud. Comparison of transported scalar PDF and
velocity-scalar PDF approaches to Delft ame III. C.R. Mcanique, 334:507516, 2006
[39] S. Subramaniam and S.B. Pope. A mixing model for turbulent reactive ows based on Euclidian
Minimum Spanning Trees. Combustion and Flame, 115:487514, 1997.
[40] H.A. Wouters, P.A. Nooren, T.W.J. Peeters, and D. Roekaerts. Simulation of a blu-bodystabilized diusion ame using second-moment closure and Monte Carlo methods. In Proceedings of the Combustion Institute, 26, pages 177185, 1996.
[41] H.A. Wouters. Lagrangian models for turbulent reacting ows. PhD thesis, Delft University of
Technology, 1998.
[42] H.A. Wouters, T.W.J. Peeters, and D. Roekaerts. Joint velocity-scalar PDF methods. In B.E.
Launder and N.D. Sandham, editors, Closure strategies for turbulent and transitional ows.
Cambridge University Press, 2001.
184
Part 4
Laser Diagnostics
185
Chapter 11
An introduction to laser diagnostics

N.J. Dam
11.1
11.1.1
Basic molecular spectroscopy

Molecular physics
In this chapter I will concentrate on the description of the internal energy levels of (small) molecules.
Molecules consist of at least two atoms, and I will start with a brief discussion on how these atoms
are held together. The available degrees of freedom can be subdivided, to a good approximation, into
1. External degrees of motion: motion of the center of mass
2. Rigid rotation (of the nuclear frame)
3. Internal vibrations (of the nuclear frame)
4. Electronic energy
The motion of the center of mass evidently does not contribute to the internal energy, and we will
neglect it further. The electrons can be separated because of their light mass: they are assumed to
instantaneously adapt to conformational changes of the nuclear frame, an approach that is called the
Born-Oppenheimer approximation. The total internal energy of a molecule can now be written as
E = T electronic + T vibration + T rotation
or T e + T v + T r ,
(11.1)
and below we will consider each contribution separately. The use of the capital letter T for the individual contributions is a relict from the past, in which they where called term values.
One remark about units. The international unit for energy, the Joule, is uncomfortably large
relative to the typical energies relevant in dealing with individual molecules. For practical reasons, a
much smaller energy unit would be convenient. Unfortunately, several of these are in use, notably the
electronvolt [eV], the wavenumber [cm1 ], and the Kelvin [K]. In optical spectroscopy, the use of the
wavenumber is common practice. For conversion factors between the various units, see Table 11.3
below.
187
188
Chapter 11 N.J. Dam
Chemical binding
Under normal circumstances, atoms are electrically neutral particles. So how do they stick together
to form molecules? The answer is to be found in the electronic structure. Atoms are build out of
a nucleus with Z protons and Z neutrons1 (Z is the atomic number), and an equal amount (Z) of
electrons. Typical sizes of nuclei are on the order of 1014 1015 m, whereas the typical size of an
atom is on the order of 1010 m. Nuclei, therefore, can be considered as point masses. In an atom,
the space around the nucleus is lled by the electrons. The spatial distribution of these electrons,
however, is not at all uniform. Rather, electrons are found in one to several2 spherical shells around
the nucleus, these shells being separated by regions of relatively low electron density. The inner shells
are closest to the nucleus, and its electrons are strongly bound. The electrons in the outermost shell,
on the other hand, are relatively far from the nucleus. Moreover, the positive charge of the nucleus is
shielded from the outer electrons to a large extent by the electrons in deeper shells. The net result
is that the electrons in the outer shells are relatively weakly bound to the atom. In fact, it is these
electrons, called valence electrons, that determine the chemical reactivity of an atom. Thus, as far as
chemical bonding is concerned, we consider an atom to consist of an atomic core made up of nucleus
plus all electrons in inner shells, and the electrons in the outermost shell. So far for atoms.
When two compatible atoms meet, the valence electrons are perturbed and rearranged over both
nuclei. The inner electrons are so closely bound to their own nucleus that, under normal circumstances, they do not notice the intruding other atom at all. When the new distribution of outer electrons
is such that the repulsion of the two positive nuclei is held in check by attraction to sucient electrons in the region between the two nuclei, a chemical bond is created. Chemical reactions between
molecules proceed in a similar way. When two molecules meet3 , their outer electrons are redistributed.
The result may be a redistribution of the atomic cores, and then we say that a reaction has taken place.
Figure 11.1: Charge distribution in the formaldehyde molecule (left) and in the water molecule (right). This is a
numerical result, and in fact shows cross sections of the change in electron charge density between the molecule
and the individual atoms in the same conguration. Blue indicates charge accumulation, red depletion. Clearly,
charge accumulates in regions between the atoms in these stable molecules.
For now, the take-home message is to realize that for a molecule to be stable, the outer electrons
must be distributed in such a way that there is sucient negative charge present between all pairs
of (positively charged) atomic cores. Molecules can then be seen as bound multi-particle systems,
1
Only the H-atom has a nucleus without neutrons, but it has isotopes (deuterium (D) and tritium (T)) that do contain
neutrons.
2
Number depends on the number of electrons.
3
You may prefer to say collide, since after all they move with considerable speed, about 300 m/s or 1500 km/h at room
temperature.
189
with relatively heavy atomic cores (at least two), and several light and relatively mobile electrons.
As we know from elementary Quantum Mechanics, the possible internal energies of bound (that is,
conned) systems are discrete, and molecules are no exception. The outer electrons can be distributed
around the atomic cores in several, discrete ways, the atomic cores can vibrate relative to each other at
only discrete frequencies, and the molecule can rotate as a whole, again only at discrete, well-dened
frequencies. The energies associated with these various degrees of freedom will be discussed below.
In the following discussion we will only treat diatomic molecules, which not only will serve to
illustrate the point, but is also of considerable interest for combustion and combustion diagnostics.
The extension to poly-atomic molecules, however, is not straightforward, and outside the scope of
this course.
Electronic energy levels
Basic features for diatomic molecules
There are no analytical expressions for the electronic part, T e , of the total energy of even the simplest
molecule. The Schr dinger equation is too complicated to be solved in closed form, but since the
o
electrons are conned to the molecule, we know that the solutions will be discrete. Also, we can
expect there to be many solutions, and this, too, turns out to be the case.
The possible electron distributions are labeled by so-called molecular term symbols, which summarize a number of symmetry properties of the molecule4 . Traditionally, the general notation of these
term symbols is as follows:
name 2S +1
(11.2)
u/g
and a few specic examples are
X 3
g
or
X 1 +
g
or
A 2 + ,
for the electronic ground states of molecular oxygen and nitrogen, and the rst electronically excited
state of the OH-radical, respectively.
The information contained in these term symbols is as follows:
name: A single letter that serves as a name. Capital X is reserved for the electronic ground state, the
rest runs alphabetically, approximately in the order of increasing energy. Capital letters denote
states with the same spin as the ground state, lower case letters denote states with other spin.
2S + 1: S is the total spin5 of the electron conguration, which can be integral or half-integral.
: A letter-code for the component of electron orbital angular momentum about the internuclear axis.
This is an integer, and the code is (for = 0), (for = 1), (for = 2), etc.
u/g: Only for homonuclear diatomics, in which it species the parity, the symmetry of the wavefunction under inversion of all coordinates.
: Species the symmetry of the wavefunction under reection through any plane containing the
internuclear axis. Both symmetries occur in states with > 0, so this property is only indicated
for -states.
4
Atomic term symbols also exist, but carry slightly dierent information.
Spin is an intrinsic property of elementary particles, like mass and charge. Spin can be seen as an intrinsic magnetic
moment. It is a vector property, the size of which is xed for a given
particle, but the orientation can vary. The magnitude
of the spin vector is given in terms of the spin quantum number s, by s(s + 1) ; s can take integral or half-integral values.
1
For electrons, protons and neutrons, s = 2 .
5
190
Chapter 11 N.J. Dam
Figure 11.2: Two typical examples of potential energy curves for a diatomic molecule (internuclear distance
r). The blue curve corresponds to a bound state, with a distinct minimum at the equilibrium distance between
the two atoms. The red curve corresponds to a dissociative state, the potential energy of which keeps on
decreasing as the atoms move away from each other. Note that both states (in this example) converge to the
same dissociation limit, that is, if the molecule breaks apart (spontaneously in the red case, external energy is
required in the blue case) the atomic products are the same in both cases. (How could they be dierent, in a
diatomic molecule?)
Within the scope of this course, the physics behind these properties necessarily has to remain obscure.
For practical purposes, it is especially important that absorption and emission of light can change
these properties only in a very restricted way, as we will see in the next chapter.
An electronic state describes a distribution of electrons. The gross features, the symmetry properties of such a distribution, are captured by the term symbols, but the exact value of the electronic
energy also depends on the distance between the nuclei. This information is typically provided using
so-called potential energy curves: plots of the electronic energy in a given distribution as a function of internuclear distance r. Examples for molecular nitrogen and molecular oxygen are shown in
Figures 11.3 and 11.4. Typically, such a plot takes one of two characteristic shapes (gure 11.2):
Potential well: There is a distinct minimum, that denes the equilibrium internuclear distance, re ,
of the molecule for that particular electronic state. If r 0 the energy goes to innity (the
positively charged atomic cores approach each other, and push away the electrons in between),
if r the curve levels o (the molecule dissociates into two atoms that behave as free
particles).
Gradual decay: The energy decreases monotonically if the nuclei move away from each other. This
does not correspond to a bound state, but rather to what is called a dissociative state.
A note on polyatomic molecules
The major aspect of polyatomic molecules is that there is no longer a single internuclear distance.
191
Figure 11.3: Potential energy curves for the nitrogen molecule, N2 . Most of the data on which these curves are
based are derived from spectroscopic experiments, and the gure is undoubtedly incomplete. A few curves that
are often encountered in practice are highlighted.
192
Chapter 11 N.J. Dam
Figure 11.4: Potential energy curves for the oxygen molecule, O2 . Most of the data on which these curves are
based are derived from spectroscopic experiments, and the gure is undoubtedly incomplete. A few curves that
are often encountered in practice are highlighted. The yellow comet tail indicates a particular feature of the
O2 molecule: the B-state is highly unstable, and decays within picoseconds to a range of states symbolized by
the yellow curve, and then into separate oxygen atoms.
193
0.5
potential energy curve

parabolic fit of minimum
0.4
y = Intercept + B1*x^1 + B2*x^

Equation
0.3
No W eighting
Energy (arb. units)
W eight
6.58579E-6
Residual Sum of
Squares
0.2
0.94378
Adj. R-Square
Value
Intercept
Standard Error
1.16145
B1
6.86767
-12.59375
2.03949
B2
0.1
5.57174
0.8943
0.0
-0.1
-0.2
-0.3
1
Internuclear distance (A)
Figure 11.5: Potential energy curve for a hypothetical diatomic molecule, and a parabolic t of the region
around the minimum. Obviously, the approximation only works well if you stay close to the energy minimum.
Thus, the potential energy plots of the diatomic molecule become multi-dimensional potential energy landscapes in polyatomic ones. Also, the symmetry properties and operations for non-linear
molecules are dierent from those for linear (incl. diatomic) molecules, and this requires a modication of the notation. Group theory is required for proper accounting of all symmetry aspects.
Vibrational energy levels
The basic model
The electronic potential energy curve of any particular electronic state determines the potential energy
of the system as a function of internuclear distance. Thus, it denes the force eld for the nuclear
motion. For values of the internuclear distance not too far away from the (classical) equilibrium value
re , the potential looks parabolic (see Figure 11.5), like that of the well-known classical system of two
particles connected by a Hookes law spring, for which
V(r) = 1 q s (r re )2 ,
2
with q s the spring constant. The approximation in which the actual potential energy curve is replaced
by the best t parabola is called the harmonic oscillator approximation. For vibrational energies
below the bond energy, this remains a bound system, and the vibrational energies are discrete. To rst
order, they are given by
T vib = e (v + 1 ) with v = 0, 1, 2, 3, . . . ,
(11.3)
2
in which e is the vibrational constant, v is called the vibrational quantum number, and the vibrational
energy is taken to be in units of wavenumbers, cm1 . The value of e is related to the curvature of the
194
Chapter 11 N.J. Dam
potential energy curve near the minimum:

e = 0.01
q s /
,
2c
(11.4)
in which the weird factor 0.01 is introduced to get e in cm1 with all other quantities in SI units.
Clearly, steeper curves have larger vibrational constants. Of course, the harmonic approximation is
just an approximation, and the above expression for the vibrational energy should be seen as kind of
a rst term in a Taylor series for the actual energies.
Renements
There are two peculiar features in the harmonic oscillator model: it predicts equidistant energy levels,
and the lowest possible energy is not zero, but 1 e . Since the real potential well gets broader towards
2
the top (relative to the harmonic approximation), the actual energy levels must get closer if the vibrational quantum number increases. This indeed happens, and a more accurate approximation to the
vibrational energy is provided by
T vib = e (v + 1 ) e xe (v + 1 )2 + e ye (v + 1 )3 + . . . ,
2
2
2
(11.5)
in which all constants are positive, and e

e xe
e ye . Usually, the products e xe and e ye are
taken as single constants. In practice, the number of bound vibrational states is nite.
1
The occurrence of a zero-point energy, the lowest vibrational energy is not zero but 2 e , is a particular quantum feature, related to Heisenbergs uncertainty relation. Although it has some interesting
philosophical consequences, it is of no concern for the daily practice of spectroscopy.

In a diatomic molecule there is only one coordinate along which vibration can occur (the internuclear
axis). More possibilities exist in larger molecules. As in classical mechanics, the internal motion
of atoms relative to each other can be described in terms of normal coordinates. In a non-linear,
N-atomic molecule there are 3N 6 internal degrees of freedom6 , i.e. 3N 6 independent normal
coordinates. Each of these is associated with an own vibrational mode, comparable to the diatomic
vibration described above, with its own vibrational constants and its own vibrational quantum number.
In a higher order approximation, cross terms accounting for mutual inuence of dierent vibrational
modes should also be taken into account.
Question
How many vibrational modes will there be for the water molecule? Can you guess what the normal
modes will look like?
Rotational energy levels
The basic model
In rst approximation, the rotational energy of a molecule is described by treating the molecule as a
rigid body: we assume the relative positions of all atomic cores to be xed, and consider them as point
masses. The rotational energy is then expressed in terms of the moments of inertia. Any rigid body
6
3N 5 in linear molecules.
195
has three principal axes, which, in symmetric molecules, coincide with symmetry axes. The principal
moments of inertia are dened relative to these principal axes:
def
I xx =
m j x2 ,
j
etc.
(11.6)
in which the summation runs over all atomic cores with masses m j at a distance x j from the principal
axis. The light electrons are neglected. In general, all three principal moments of inertia are non-zero.
An important exception to the latter is the case of linear molecules, including all diatomics: since we
consider the atomic cores as point masses, the moment of inertia for rotation around the internuclear
axis is zero. Thus, there is no energy associated with rotation around the internuclear axis, and linear
molecules have only two principal axes of rotation, both perpendicular to the internuclear axis.
The expression for the rotational energy depends on the molecular symmetry, and we will treat
here only the case for linear molecules (that is, including diatomics).
Linear molecules:
T rot = BJ(J + 1)
with
J = 0, 1, 2, 3, . . .
(11.7)
In this equation, the constant B is called the rotational constant, and is dened by
def
B =
2I
or B [cm1 ] =
0.01
,
4cI
(11.8)
and I is the moment of inertia relative to any axis perpendicular to the internuclear axis. The
form between parentheses holds when B is to be expressed in wavenumbers.
As with vibration, this simple expression can be seen as the rst term in a Taylor series expansion.
For non-linear molecules the expression is more complicated (symmetric molecules) or non-analytical
(asymmetric molecules).
Exceptions and renements
Two obvious improvements over the rigid-body assumption are to account for centrifugal stretching
and for the dependence on vibration. Centrifugal forces cause bonds to stretch, so that the moment
of inertia increases and the B-constant decreases. Vibration causes bonds to periodically stretch and
compress, but since the distance occurs squared in the expression for the moment of inertia, the Bconstant eectively decreases with increasing vibrational amplitude. Taking these eects into account
to rst order produces the following expression for the rotational energy levels:
T rot = (Be e (v + 1 ))J(J + 1) De J 2 (J + 1)2
2
(11.9)
involving two new (positive) constants e and De . All constants depend on the electronic state the
molecule is in (because that determines the equilibrium distances), and that has been explicitly indicated by the subscripts e in the above equation.
In diatomic molecules in which the spin quantum number is half-integral (e.g. NO, which has
2S +1 = 2 S = 1 in the ground state), there is an additional electronic contribution to the rotational
2
energy, and the rotational quantum number becomes half-integral as well. We will not discuss this
further.
In a diatomic molecule (and linear molecules in general), one principal moment of inertia is zero (for
196
Chapter 11 N.J. Dam
rotation about the internuclear axis), and the other two are equal. In larger molecules, the three principal moments may all be equal (in so-called spherical top molecules, like methane, CH4 , and SF6 ),
two may be equal and one dierent but not zero (symmetrical top molecules, like ammonia, NH3 ,
and methanol, CH3 OH), or all may be dierent (asymmetric top molecules, like water, H2 O). The
number of rotational constants and rotational quantum numbers is equal to the number of unique moments of inertia. Spherical molecules behave like linear ones, symmetrical tops obey a slightly more
complicated equation with two quantum numbers, and for asymmetric top molecules no analytical
expressions exist.
Summary expressions for the internal energy of diatomics
The internal energy levels of diatomic molecules are labeled by three quantum numbers: the name
of the electronic state, the vibrational quantum number v, and the rotational quantum number J. In
zeroth order:
E E(e, v, J) = T e + T v + T r = T e + e (v + 1 ) + Be J(J + 1) .
(11.10)
2
For T e the minimum of the potential energy curve for the electron conguration at hand is taken. The
energy zero is often chosen at the minimum of the ground electronic state potential energy curve (I
will do so, too), but sometimes it includes the zero-point energy of the ground state as well.
To a better approximation, more accurate for larger values of the quantum numbers, one can write
E E(e, v, J) = T e + T v + T r
1
= T e + e (v + 2 ) e xe (v + 1 )2 + Bv J(J + 1) Dv J 2 (J + 1)2 ,
2
(11.11)
with
Bv = Be e (v + 1 ) ,
2
and a similar expression for Dv . For a number of diatomic molecules relevant to combustion and/or
combustion diagnostics, the spectroscopic constants are listed in Table 11.1. More constants can be
found in the Chemistry Webbook of NIST, http://webbook.nist.gov/chemistry/ .
There is a distinct hierarchy in the numerical values of the various energy terms. Apart from the
ground electronic and vibrational states (for which T e , T v = 0), we typically have
Te
Tv
Tr .
Thus, the coarse energy level structure is determined by the electronic states. The vibrational states
provide ner structure to each of the electronic states, and each vibrational state is further subdivided
into rotational states. Figure 11.6 shows as an example the potential energy curve of the ground state
of the nitrogen molecule, in which the vibrational and rotational states are indicated.
The Boltzmann distribution
Now that we have found expressions for the possible internal energies of molecules (or at least for
diatomic ones), we may ask in which state we would nd any particular molecule if we measured it.
The answer must be that we do not know. A common misconception is that molecules would strive to
achieve a state of minimum energy: this is incorrect for several reasons. What can be said, however,
is that in an ensemble of identical molecules that is in thermal equilibrium, the number of molecules
68152.66
50203.6
0
49793.28
7918.1
0
20022.50
8145
716.2
0
100089.8
91700.0
0
69774
32684.1
126.23
0
31801.5
23189.8
0
65075.7
55825.4
48686.70
0
53084.7
43965.7
119.82
0
a 1
u
A 3 +
u
X 1 +
g
B 3
u
a 1 g
X 3
g
d 3 g
B 1 u
a 3 u
X 1 +
g
C 1 u
B 1 +
u
X 1 +
g
B 2 +
A 2 +
X 2 1/2
X 2 3/2
C 2 +
A 2
X 2
A 1
a 3 +
a 3
X 1 +
D 2 +
A 2 +
X 2 3/2
X 2 1/2
N2
NO
CO
CH
OH
H2
C2
O2
Te
[cm1 ]
Electronic
state
Molecule
1530.25
1460.64
2358.57
709.31
1483.50
1580.193
1788.22
1608.2
1641.3
1854.71
2443.77
1358.09
4401.21
660.0
3178.8
3737.76
3737.76
2840.2
2930.7
2858.5
1518.2
1228.60
1743.4
2169.81358
2323.90
1017
1904.04
1904.20
e
[cm1 ]
92.91
84.881
84.881
125.96
96.65
63.02
19.40
10.468
14.3
13.28831
22.885
11
14.100
14.075
12.0747
13.87
14.324
10.65
12.9
11.981
16.440
12.1
11.7
13.340
69.524
20.888
121.33
e xe
[cm1 ]
1.4799
1.4546
1.99824
0.81902
1.4264
1.4376766
1.7527
1.62
1.63
1.81984
31.362
20.015
60.853
5.086
17.358
18.910
18.910
14.603
14.934
14.457
1.6115
1.3446
1.69124
1.93128087
2.0026
1.9965
1.72016
1.67195
Be
[cm1 ]
0.786
0.7242
0.7242
0.7185
0.697
0.534
0.0232
0.0189
0.01904
00.01750441
0.02175
0.01915
0.0182
0.0171
0.01661
0.01765
1.664
1.1845
3.062
0.01657
0.0180
0.017318
0.01206
0.0171
0.01593
0.01608
e
[cm1 ]
6.44
6.92
2.23
1.625
4.71
929
20.39
19.38
19.38
1555
1540
1450
7.33
6.41
6.36
6.1214
5.8
5.4
10.2
0.5
5.55
6.1
5.76
4.55
4.86
4.839
6.7
De
cm1 ]
[106
1.31190
1.24253
1.0327
1.2928
0.74144
1.869
1.0121
0.96966
0.96966
1.1143
1.1019
1.1199
1.2353
1.3523
1.20574
1.128323
1.0618
1.0634
1.15077
1.15077
1.2755
1.2866
1.09768
1.60426
1.21563
1.20752
1.2661
re
[]
Table 11.1: Spectroscopic constants for some diatomic molecules relevant for combustion or combustion diagnostics, up to the last signicant digit. Note the
remarkable accuracy of some parameters!

197
198
Chapter 11 N.J. Dam
Table 11.2: Values of some fundamental constants, all in SI units. (Source: 2006 CODATA recommended
values. See the Fundamental Constants website, at http://physics.nist.gov/cuu/index.html .)
Constant
symbol
speed of light
magnetic constant
electric constant
Planck constant
h/2
elementary charge
atomic mass unit
electron mass
electron/proton mass ratio
Rydberg constant
Avogadro constant
Boltzmann constant
Stefan-Boltzmann constant
CODATA value
2.99 792
c
0
458108
4107
h
e
u
me
m p /me
R
NA
kB
1/0 c
6.626068961034
1.0545716281034
1.6021764871019
1.6605387821027
9.109382151031
1.83615267247103
1.0973731568527107
6.022141791023
1.38065041023
5.670400108
Daily use value

3108
4107
8.91012
6.61034
11034
1.61019
1.661027
0.911030
1.84103
1.1107
61023
1.381023
5.67108
SI unit
m s1
N A2
F m1
Js
Js
C
kg
kg
m1
mol1
J K1
W m2 K4
Table 11.3: Conversion between energy units. The various energy units are based on the following relations
E = h = hc/ = hc = kB T = qe V. The last expression is used in the denition of the electronvolt, eV: the
energy gained by an electron upon acceleration by a voltage of 1 Volt. It would seem logical to express the
wavenumber, = E/hc, in units of m1 . For historical reasons one uses cm1 . In chemistry one still often
uses the archaic kcal/mol as a unit. In order to speed up the extinction of this unit, I will not even provide a
conversion factor here. The table below can be used for quick interconversion between the various units. To
convert e.g. J to eV, select the value in the cell in row J and column eV: 1 J 6.24 1018 eV.
J
eV
cm1
K
eV
cm1
6.24 1018
5 1022
8066
7.2 1022
11556
1.439
1019
1.6
2 1023
1.38 1023
1.24 104
8.6 105
0.695
199
Figure 11.6: Electronic, vibrational and rotational energy levels in the nitrogen molecule, N2 . Energy along
the vertical axis, drawn to scale with the magnication factors as indicated. Left: Potential energy curve of the
ground state (X) and of a small part of the lowest electronically excited state (A). Middle: Lowest vibrational
energy levels in the ground electronic state, in the harmonic oscillator approximation (left) and with the rst
order correction (the e xe -term; right). Right: Lowest few rotational energy levels in the vibrational ground
state of the electronic ground state, in the rigid rotor approximation (left) and with the rst order centrifugal
correction (the De -term; right). Note that on the scale of this drawing, the rst-order correction on the vibration
is barely visible, and that on the rotation not at all. With modern lasers, however, the eect is easily measurable.
per available quantum state obeys Boltzmann statistics. Thus, we can make a statement about the
average number of molecules that will be in a particular energy state in a system at temperature T :
gevJ exp{E(e, v, J)/kB T }
N0
Z(T )
E(e, v, J)
with Z(T ) =
gevJ exp{
}.
kB T
evJ
N(e, v, J) =
(11.12)
(11.13)
N0 is the total number of molecules (or the number density), Z(T ) is called the partition function, kB
is the Boltzmann constant, and gevJ is the degeneracy factor of each energy level, that, in diatomic
molecules, typically depends on J only7 and is then given by gevJ g J = 2J + 1. The quantity
N(e, v, J) is called the population of the level characterized by the quantum numbers e, v, J, and is
typically expressed in a percentage of the total number of molecules, N0 . The fraction N(e, v, J)/N0 is
called the Boltzmann fraction. The population dierence between any two states can most easily be
expressed by the population ratio, because then the partition function cancels:
N
N(e , v , J )
E
g
=
=
exp{
},
N
N(e , v , J ) g
kB T
(11.14)
in which
E = E(e , v , J ) E(e , v , J ) .
7
In homonuclear diatomics there is another contribution that depends on nuclear spin. It is an important eect, so you
should be aware of its existence, but we will not consider it here.
200
Chapter 11 N.J. Dam
In the approximation that electronic, vibrational and rotational energies are completely independent, the partition function can be factorized:
gevJ exp{
E(e, v, J)
}
kB T
gevJ exp{
Z(T ) =
Ee + Ev + Er
}
kB T
evJ
=
evJ
exp{
e
Ee
}
kB T
exp{
v
Ev
}
kB T
g J exp{
J
EJ
} = Zel Zvib Zrot .
kB T
(11.15)
With this, we can more easily calculate populations of electronic, vibrational or rotational states separately, for instance
exp{Ev /kB T }
,
Nv = N0
Zvib (T )
and similarly for electronic or rotational population distributions. Figure 11.7 shows an example of the
population distribution in the N2 molecule at room temperature and at 2000 K, what typically might
be encountered in a ame. As an exercise, you might try to reproduce the numbers in this gure. (Use
the constants of Table 11.1.)
12
100
100
80
80
60
40
60
40
20
20
Population (in %)
Population (in %)
P opulation (in % )
10
A
Electronic state
8
6
4
2
10
15
20
25
30
35
40
45
50
Vibrational quantum number
Rotational quantum number
Figure 11.7: Population distribution over electronic (left), vibrational (middle) and rotational (right) states in
the nitrogen molecule, N2 , in thermal equilibrium at room temperature (300 K, black) and at a typical ame
temperature (2000 K, red). The height of each bar corresponds to the population of the state labeled by the
quantum number below the bar pair. The rotational population alternation is a result of nuclear spin statistics:
in N2 the levels with odd J have an extra degeneracy factor of 2.
From a microscopic point of view, a gas is a most violent environment. Even at STP, molecules
y around at random with the speed of sound, and experience collisions with other molecules at a
rate of about 109 per second. These collisions mediate a permanent exchange of energy between
collision partners, and when we speak of thermal equilibrium it is a highly dynamical equilibrium
that we actually mean. The thermal energy available in gas phase collisions is kB T on average. At
room temperature, this corresponds to about 200 cm1 of energy, far too little to excite (almost all)
electronic states, and even too little to excite many vibrational states. Thus, under conditions of room
temperature (or below), you can typically consider all molecules in a gas to be in their ground electronic and vibrational (vibronic) state. Even in a ame (2200 K 1500 cm1 ), electronic states are
201
not populated, but vibrational states may carry considerable population. In all cases, many rotational
states are typically populated. (A particular state is said to be populated if, on average, a signicant8
fraction of the molecules is found in that state.)
In polyatomic molecules, especially those with heavier atoms, there may be vibrational states at low
energy9 . These states, and perhaps combinations thereof, then need to be taken into account for the
partition function. When the molecules possess symmetry, vibrational states may be degenerate, and
the variation of degeneracy over combinations of vibrational states is a tricky business. This, again, is
a topic we will not deal with here.
11.1.2
Spectroscopy
Spectroscopy, according to Wikipedia the ultimate source of perpetual wisdom , was originally
the study of the interaction between visible radiation and matter as a function of wavelength of
the radiation. This is still a good working denition, if visible is replaced by electro-magnetic.
Arguably, spectroscopy is the most powerful tool that we have for the study of molecular structure.
The huge amount of available knowledge also allows spectroscopy to be used as an analytical tool
to study the chemical composition of unknown mixtures; we often speak of spectrometry in such
applications.
Given that we are interested primarily in combustion, and will use laser diagnostics as a spectrometric tool, I will focus on spectroscopy of the gas phase. To a good approximation, this can be
described in terms of the interaction of light with individual molecules10 . Molecules always have to be
described quantum mechanically (cf. chapter 11.1.1), but for the light one can often choose between
a wave (radiation) or particle (photon) description, whichever is more convenient. Only a relatively
small amount of interactions really requires the more fundamental photon picture to be used, and in
those cases we speak of quantum optics.
I have split our discussion of spectroscopy, somewhat articially, in two parts: wavelength dependence, and interaction strength. This is preceded by a brief discussion of radiation physics. One
remark (or warning) at the start: spectroscopy is an old science. This has the advantage that much
knowledge is already available. A disadvantage is that the notation is often xed by tradition rather
than by logic.
Fundamental principles of spectroscopy
Basic photometric quantities
We will need to quantify the energy exchange between a light beam and a gas. Here I will briey
review the options. Units (all in the international SI system) are added in [square brackets].
The total amount of energy W emitted by a light source and falling on a particular surface is called
the radiant energy [J]. The radiant energy per unit time is called the radiant power or radiant ux [J/s
W]; it is often denoted by the letter P. The radiant energy density is the radiant energy per unit
volume [J/m3 ], often denoted . The word radiant, by the way, is often omitted in daily use.
8
Should you think that signicant is not a particularly well dened number, then you are right.
Low, that is, comparable to ambient temperature.
10
In liquids and solids the density is so high that molecules tend to loose their individuality, and collective eects become
important.
9
202
Chapter 11 N.J. Dam
Figure 11.8: The two-level molecule as a photophysical prototype, with various processes that may occur under
irradiation.
All these quantities are integrated over all wavelengths. They all have spectral counterparts, dened by the quantity per unit frequency. For instance, the spectral radiant power P is dened as
the radiant energy per unit time per unit frequency [J/s/Hz or W/Hz]. The spectral and integrated
quantities are related in the obvious way, e.g.
P=
P () d .
0
Sometimes, the spectral unit is taken dierently, for instance in wavenumbers, , or even in terms of
wavelength, = c/. The conversion follows

P ( ) = 3 1010 [Hz/cm1 ] P ()

and
P () =
c
P () .
2
Finally, the radiant power incident on unit surface area is called the intensity, or perhaps better irradiance [W/m2 ], typically denoted by I. This, too, has a spectral counterpart. Irradiance and radiant
energy density are related by
I = c .
Radiation is often described in terms of its electric eld strength, and in terms of plane waves. A
plane wave is a wave of which the amplitude varies in one direction only, e.g.
E(z, t) = E0 cos(t k r) ,
which, at a given moment in time, only varies along k, but not in planes perpendicular to k. (Why
is that?) This is also an example of a so-called monochromatic wave, since only one colour (that is,
wavelength) is present in the eld. Through Fourier analysis, any eld can be expressed in terms of
plane waves, so the restriction in the description to only plane waves is no real limitation. For such
plane, monochromatic waves traveling along the z-axis, the irradiance is related to the eld strength
by
2
1
I(z) = 2 0 cE0 .
Basic photophysical processes
Consider as an example a gas of very simple molecules, with only two energy levels, as in gure 11.8,
which is illuminated by white light of energy density (). Three processes may occur:
Induced absorption: Molecules in the lower state can absorb a photon and make a transition to the
upper state. The probability of this happening is proportional to the energy density of the light
eld, and the density of molecules in the lower level.
203
Induced emission: Molecules in the upper state can be stimulated to emit a photon and make a
transition to the lower state. This is the reverse process of the previous one, and the emitted
photon indeed has the same properties (frequency, direction, polarization) as the incident one.
This process, too, is proportional to the elds energy density, and to the number of molecules
in the upper level.
Spontaneous emission: Molecules in the upper level may decay spontaneously to the lower state,
under emission of a photon. This process is independent of the external eld, and the emitted
photon has random direction and polarization.
The second-mentioned process is perhaps something that you hadnt expected: by irradiating a medium,
you gain light rather than loose it. Quantum mechanics predicts that light can induce transitions between energy levels. It does not, however, predict a direction of this transition; upwards or downwards
is equivalent. The direction is determined simply by the initial level, the level that the molecule is in
when you switch on the light. If it is in the upper state, it can only go down; if it is in the lower
state, it can only go up. In practice, the Boltzmann distribution of thermal equilibrium (see above,
in particular Figure 11.7) typically leaves the upper level of the transition of interest with negligible
population. Chemical reactions, like those in combustion, may create reaction products in excited
states, which then may decay to the ground state by spontaneous emission of light. The blue colour
of a gas ame, for instance, is due to spontaneous emission of photons by excited-state CH-radicals
produced in the reaction zone of the ame. The laser is based on the process of stimulated emission,
as we will discuss later in the course.
Spontaneous emission causes population to leak back from the upper state into the lower state.
The 1/e decay time of the upper level is called its natural lifetime, 2 , and is given by
2 =
1
.
A21
(11.16)
The proportionality constants B12 , B21 and A21 that determine the probability of each of the three
processes of gure 11.8 are called Einstein coecients, after the one who introduced this scheme.
They are not independent, but obey the following relations:
g1 B12 = g2 B21
and
A21 =
3
21
2 c3
B21 ,
(11.17)
in which the g-factors denote the level degeneracies and 21 is the angular frequency corresponding
to the energy dierence between the two levels, 21 = E2 E1 . The Einstein coecients can be used
to describe the evolution of level populations under the inuence of incident irradiation by means of
rate equations. For the specic case of Figure 11.8 these read
N1 = B12 ()N1 + B21 ()N2 + A21 N2
N2 = +B12 ()N1 B21 ()N2 A21 N2 = N1 .
(11.18)
(11.19)
Obviously, the Einstein B12 coecient will have something to do with the absorption of light by
a medium that is composed of these specic molecules. In order to nd this relation, we must rst
introduce a denition of the absorption coecient.
The basic experiment
For the sake of deniteness, consider the experiment sketched in gure 11.9. The output of a light
204
Chapter 11 N.J. Dam
Figure 11.9: Basic experimental setup to study the absorption of light by a gas.
Figure 11.10: Part of the absorption
spectrum of the NH radical, measured
in the ultra-violet spectral range. In
terms of eq. 11.21, the transmission is
dened as P(z = L)/P(z = 0).
1.0
Transmission
0.8
0.6
0.4
0.2
0.0
320
325
330
335
340
345
350
Wavelength (nm)
source is sent through some kind of spectral lter, that passes only one single (but adjustable) frequency (or wavelength, if you prefer). Thus, this combination creates a monochromatic light beam.
This beam traverses a cell (with optical windows on both ends) that can be lled with the gas of
interest, and the transmitted power is measured. The question would be: how does the transmitted
power vary with the frequency of the incident light? In practice, you should realize that such variation
can have various causes: the output of the light source can vary with wavelength, the lter eciency
might depend on wavelength, the cell windows may be more transparent for some wavelengths than
for others, and even the response of the power meter may depend on wavelength. We are, however,
only interested in the absorption of light by the gas in the cell, so we will assume that everything else
is constant. In the regime of linear absorption, that is, when the light intensity is suciently low that
the properties of the gas are not aected by the presence of the light, we can use Lambert-Beers law
to describe the absorption. This law states that the spectral power loss over any innitesimally small
path is proportional to the incident spectral power and the path length:
def
P (z) = P (z + dz) P (z) = (z, )P (z) dz
(11.20)
The proportionality constant (z, ) is called the local absorption coecient. Its spatial dependence is
governed by the spatial distribution of the gas and perhaps the temperature, and its spectral dependence
is determined by the chemical composition of the gas. This can be formally integrated over the whole
cell length to yield the power at the end of the cell:
L
P (z = L) = P (z = 0) exp{
}(z, ) dz
0
(11.21)
205
When the absorption coecient is uniform, this expression simplies to

P (z = L) = P (z = 0) exp{()L}
(11.22)
Note, by the way, that can just as well be taken as a function of wavelength rather than frequency;
it is for consistency with other sections that I chose frequency here. A measurement of the light
power transmitted by the cell then might look as in gure 11.10. Such a graph is called an absorption
spectrum. Each chemical species has a characteristic absorption spectrum, and, at least in principle,
from the accurate measurement of an absorption spectrum the composition of the absorbing gas can
be derived11 .
The absorption coecient introduced above is not a fundamental material property, since it obviously increases if the gas density increases. We write it as
(z, ) = ()(N (z) N (z)) ,
(11.23)
in which N and N describe the number densities12 [molecules/m3 ] of molecules in the lower and
upper states, respectively, of the transition involved, and () is the frequency-dependent absorption
cross section of the transition [m2 /Hz]. The latter can be seen as the geometrical surface area of
the molecule with which it removes light from an incident beam. We will return to its frequencydependence later in this chapter, when we discuss lineshapes. If the upper level is not populated,
N (z) = 0 and (z, ) = ()N (z).
Wavelength dependence
Light can be pictured either as a wave, or as an ensemble of photons. The latter is a bit more convenient
in getting a basic understanding of what goes on. The two pictures are connected through energy:
photons are massless particles that carry an energy13 which is proportional to the frequency of the
radiation:
hc
E = h = =
= hc = ck ,
(11.24)
in which h is Plancks constant, is the frequency, = h/2, = 2 is the angular frequency, c is

the speed of light, is the wavelength, and = 1/ and k = 2/ are two forms of the wavenumber.
On a microscopic level, spectroscopy involves the interaction of individual photons and molecules.
In the interaction, photons can be absorbed (light absorption) or emitted (light emission). The law
of conservation of energy then requires this energy to be accommodated by the molecule with which
the interaction takes place. Thus, this molecule gains energy (photon absorption) or looses energy
(photon emission). The crucial point is now that, since the allowed energy levels of molecules are
discrete, also the amounts of energy uptake and release must be discrete. Both before and after the
interaction, a molecule must be in an allowed energy state. Thus,
Ef = Ei h ,
(11.25)
in which the subscripts indicate nal and initial molecular energy states, h is the photon energy, and
the + or sign should be used for light absorption or emission, respectively. In publications, you
11
In practice, performing a proper absorption experiment is not easy. Major reasons are poor sensitivity and spectral
interference.
12
Not the mass densities.
13
Photons also carry momentum, according to the de Broglie relation, p = h/. See the section on transition lineshapes,
below.
206
Chapter 11 N.J. Dam
often nd another (traditional) notation, with double primes labeling the lower energy state (e.g. E )
and single primes labeling the upper energy state. The above equation then reads
E E = h =
hc
= hc ,
irrespective of whether you deal with absorption or emission. The summary equations 11.10 and 11.11
of the previous chapter can now be used to determine at which wavelengths absorption or emission is
to be expected:
hc
E(e , v , J ) E(e , v , J ) = h =
= hc .
For the simplest case of eq. 11.10 it reads

1
= (T e T e ) + (e (v + 2 ) e (v + 1 )) + (Be J (J + 1) Be J (J + 1)
= Eelec + Evib + Erot .
(11.26)
Note again that this notation assumes all spectroscopic constants to be given in wavenumbers. (So on
the left-hand-side we have h/hc = /c = 1/ = .)
The large dierence in energy interval between electronic states, vibrational states and rotational
states usually determines the spectral range in which transitions can be found. In practice, and certainly in absorption experiments in which the initial level is in the ground electronic state, a transition
between electronic states (irrespective of concomitant transitions between rotational and/or vibrational states) involves light in the visible or ultra-violet range of the spectrum. Transitions between
vibrational states (no change of electronic state, but there may be a concomitant change in rotational
state) typically fall in the infra-red spectral range, and purely rotational transitions (only the rotational
quantum number changes) occur in the microwave region.
Transition lineshapes
From the emphasis that I put on the energy conservation law, you may have been left with the impression that the spectral location (in terms of frequency or wavelength) of any specic transition is
carved in stone: the energy levels of any molecule are discrete, and so is their dierence, and therefore absorption and emission of light can only occur at innitely well-dened wavelengths. Although
this idea is correct in theory, there are several factors that endow transitions with a nite linewidth
in practice. Some of these factors have to do with the limited resolution of measurement equipment,
others are more fundamental and caused, for instance, by thermal motion (Doppler broadening) or
intermolecular collisions (pressure broadening).
Natural linewidth
There is a fundamental reason why the linewidth of any transition is nite, and that is the nite lifetime
of at least the upper state. This lifetime is limited due to spontaneous emission, eq. 11.16. Spectral and
temporal properties are related: they are a Fourier-transform pair. Therefore, nite lifetime implies
nite spectral bandwidth. It can be shown that the nite lifetime due to probabilistic processes, like
spontaneous emission, results in a frequency distribution that has a Lorentzian functional shape. The
absorption coecient can then be written as
() = 0 F L () ,
(11.27)
207
in which F L () is called the Lorentzian line shape function,

1
.
2 ( 0 )2 + ( 1 )2
2
F L () =
(11.28)
This is a normalized prole, that is,
F L () d = 1 ,
0
with full width = 1/1 + 1/2 at half the maximum height, and 1,2 denote the lifetimes of the two
levels involved in the transition. This represents the lower limit of the linewidth of any transition14 .
Doppler broadening
The Doppler eect is undoubtedly familiar to you, but possibly only from acoustics: the alarm of an
ambulance rushing towards you sounds at a higher pitch than the signal of one that moves away from
you. This same eect plays a role in the case of light emission/absorption as well. Although the speed
of light emitted by a moving source is still the speed of light, the frequency of that light does depend
on the motion of the source. The apparent frequency perceived by an observer is given by15
= 0 (1 v/c) ,
(11.29)
in which 0 is the frequency of the light in the rest frame of the source, v is the relative velocity of
observer and source (taken positive if they move away from each other), and c is the speed of light.
Molecules of mass m in a gas in thermal equilibrium move at random, with a gaussian velocity
distribution, known as the Maxwell distribution:
1
m(v v0 )2
2
P(v) d3 v = C exp
(11.30)
.
kB T
This equation describes the probability of nding molecules with a velocity vector in the interval
[v, v + dv]. C is a proportionality constant (can be determined from normalization) and v0 is a possible
macroscopic ow velocity. Because the perceived frequency is linear in the velocity, the distribution
of perceived frequencies takes on the same functional form. And because the absorption coecient is
directly proportional to the number of molecules that absorb at a particular wavelength, the absorption
coecient, too, takes on the same form:
() = 0 FG () ,
(11.31)
in which FG () is called the Gaussian line shape or Doppler line shape function:
1
( 0 )2
FG () =
exp
22
2
with = 0
kB T
.
mc2
(11.32)
The width parameter is related to the full width, , of the prole at half the maximum height
(FWHM):
= 8 ln 2 .
(11.33)
14
Recent quantum-optical experiments have demonstrated that even the natural linewidth can be beaten, involving a
fascinating role for the vacuum.
15
There are higher-order correction terms, but these will not worry us here.
208
Chapter 11 N.J. Dam
1.0
Figure 11.11: Gaussian (Doppler) line shape and

Lorentzian line shape, drawn with the same full
width at half the maximum height (FWHM). Both
proles are normalized. The horizontal axis is in
reduced frequency units, ( 0 )/( or ).
Lorentzian
Gaussian
Line shape function
0.8
0.6
0.4
0.2
0.0
-8
-6
-4
-2
Reduced frequency
Doppler broadening becomes more important if you go to shorter wavelengths and/or to higher temperatures. A typical order of magnitude for transitions in the visible spectral range is a few GHz,
easily resolvable with modern lasers.
Pressure broadening
Measurements are hardly, if ever, performed on single, isolated molecules. Rather, you might deal
with a gas at STP16 , in which the density is about 2.5 1025 molecules/m3 and each molecule experiences about 109 collisions per second with other molecules. Such an environment must have an eect.
Indeed, during collisions, even when they are elastic, the energy levels of the participating molecules
are randomly shaken up. Eectively, therefore, the position of energy levels of molecules in a gas
is not well-dened: the environment cannot be neglected, and energy levels uctuate around some
average value.
To a good approximation, the distribution of energy levels around the average takes the form of a
Lorentzian function, like the natural linewidth discussed above (Eq. 11.28). This, then, also holds for
the absorption coecient, and we write (in the absence of Doppler broadening)
() = 0 F L () ,
(11.34)
in which the width of the Lorentzian line prole depends on the molecule and on the states involved in
the transition. Typically, it scales linearly with pressure, and under conditions of STP it is comparable
in size to the Doppler component.
Figure 11.11 compares the Doppler and Lorentz proles. It can be seen that the Lorentz prole
has more extended wings and, therefore, a lower central value than the Doppler prole. When both
(and possible other contributions to the line shape) are of comparable magnitude, the net line shape
is a convolution of the contributing components. For such a composite prole, known as Voigt line
shape prole, no simple analytical formulae exist.
Power broadening
Power broadening occurs when the incident light is suciently intense to modify the equilibrium level
populations signicantly. Essentially, what happens is that the absorption in the line center saturates
(that is, it does not increase any more with increasing incident intensity), because the level populations
Standard conditions for Temperature and Pressure, that is, 0 C and 100 kPa. Unfortunately, this standard is not really
standard, and other conventions exist.
16
209
Figure 11.12: The eect of broadening parameters on the appearance of a spectrum. Top: Absorption spectrum
around 310 nm of the OH radical in an atmospheric-pressure ame at 1500 K. The indicated part is magnied
below. a) Minimal broadening; b) Only pressure broadening ( = 76 kHz/Pa); c) Only Doppler broadening;
d) Only instrumental broadening of 0.02 nm. All spectra calculated by LIFBASE. The unreadable wavelength
scale (horizontal) is spans the range of 300320 ngstr m.
o
become equal, whereas that in the wings of the prole continues to rise. Thus, the FWHM becomes
dependent on the incident intensity. We will not discuss this eect any further.
Instrumental broadening
The nite resolution of experimental equipment can take on any shape or value, depending on the
equipment and its settings. When a laser is used as a light source, the frequency bandwidth of the
radiation is often small17 relative to the Doppler and pressure broadening of the experiment. In emission spectra, however, the resolution is typically derived from some kind of spectrometer, and these
typically have (much) less resolution. Figure 11.12 shows a few examples of the eect that resolution
has on the appearance of spectra.
Combinations of line broadening mechanisms
Typically, several line broadening mechanisms contribute simultaneously to the observed width of a
transition. In atmospheric ames, for instance, it is likely that the observed linewidth will be determined by both Doppler broadening and pressure broadening. In general, the eective lineshape is a
convolution of all individual ones. Analytical expressions for the eective linewidth only exist if all
17
But not necessarily negligibly small.
210
Chapter 11 N.J. Dam
individual broadening mechanisms result in the same line prole. We have:

1. The convolution of two Lorentzian proles is again a Lorentzian. The widths simply add up:
e = 1 + 2 .
2. The convolution of two Gaussian proles is again Gaussian. The widths add quadratically:
e =
2 + 2 .
2
1
3. The convolution of a Lorentzian with a Gaussian lineshape is called a Voigt prole. There is no
analytical expression for the eective width.
4. Any other combination results in a mess.
Some (numerical) examples are shown in Figure 11.12.
Selection rules and transition strength
The above relations predicts at which wavelengths a light-induced transition would obey the energy
conservation law. Whether or not such a transition (absorption or emission) would actually take place,
or, more generally, how likely it is that such a transition would occur will be addressed here. For the
sake of deniteness, let us focus on absorption. The probability (per unit time) that a transition occurs
is determined in two steps:
1. In symmetric molecules, symmetry restrictions impose rigid requirements on the quantum numbers of the states that can be connected by (light-induced) transitions. Transitions that obey
these so-called selection rules are called allowed transitions, those that do not are called forbidden transitions.
2. The actual value of the probability for allowed transitions to occur depends on details of the
wavefunctions of the states involved. Although fairly general expressions can be derived for the
linear molecules that we discuss here, this falls outside the scope of this course.
Selection rules allow a rst screening of which transitions can in principle occur. They exist for
all quantum numbers, electronic, vibrational, and rotational, and the details depend on the kind of
transition that is considered. Below I state the selection rules for a quantum number q in terms of
q q q . (Refer to chapter 11.1.1 for the quantum numbers.)
Upper and lower state in dierent electronic states: In this case the following selection rules hold:
Spin: S = 0
: = 0, 1
u/g: u g
: + + or
v: no restrictions (in principle)
J: J = 0 (not if = 0), or 1.
Upper and lower state in the same electronic state: In this case the selection rules can be much
more restrictive:
Homonuclear diatomic molecules: no transitions allowed at all!
211
For other molecules (heteronuclear diatomics or polyatomic molecules) we have:

v: v = 0, 1
J: J = 0 (not if v = 0), or 1 .
Some notes:
Depending on the quantum numbers that change during a transition, one speaks of electronic,
vibrational, vibronic, rovibrational, etc. transitions.
More nomenclature, probably derived from the good old days in which spectra were recorded
on photographic plates. All rovibrational transitions within a particular electronic transition
are collectively called an (electronic) band system. All rotational transitions within a particular
vibrational transitions are collectively called a (vibrational) band. An individual rotational
transition is called a line.
There is a special, again traditional nomenclature for transitions with particular values for J =
J J . Transitions with J = 1, 0, +1 are called P-, Q-, and R-transitions, respectively. In a
spectrum, all P-lines of a particular vibronic transition are collectively called the P-branch, and
equivalently we speak of Q-branch and R-branch.
The peculiar J selection rule may give rise to a peculiar shape of vibrational band envelopes18 .
In rst order (Eq. 11.26), the P- and R- line frequencies are given by (J J )
(B + B )J + (B B )J 2 (P-branch)
e
e
e
e
(11.35)
h(J) = E(e, v)
+(3B B )J + (B B )J 2 + 2B (R-branch)
e
e
e
e
e
Since in most cases Be Be , the quadratic term is typically much smaller than the linear term.
Thus, P-branches run to the red, and R-branches to the blue. Nevertheless, depending on the
sign of B = Be Be , the lines in either the P- or the R-branch will grow ever closer together
with increasing J, and there will be a value of J for which they pile up and the branch reverses
direction. This is called a band head.
In vibronic transitions there are no strict selection rules for v. That does not imply, however,
that all v are equally probable. The transition strength is determined by so-called FranckCondon factors, which represent overlap integrals of the wavefunctions of the two vibrational
states involved.
The above selection rules are based on the interaction of only the electric eld component of
the incident light with the electronic dipole moment of the molecules that make up the medium.
This is typically the strongest interaction, and that makes these selection rules relevant. There
are, however, other types of interaction, e.g. magnetic dipole or electric quadrupole interactions,
and these have their own selection rules.
Selection rules tell you whether a transition can or cannot occur based on arguments of molecular
symmetry. What they do not tell you is how strong an allowed transition will actually be. In other
words, they will predict for which transitions the absorption coecient equals zero, but they will
18
Envelope: the gross outline, not paying attention to individual rotational lines.
212
Chapter 11 N.J. Dam
not predict the actual value of for the rest. Let us write the frequency-dependent absorption cross
section of Eq. 11.23 as
() = 0 g() ,
that is, as a product of a frequency-integrated absorption cross section and a line shape function. We
then have
hB12 1 g2 2
= 4 A21 .
(11.36)
0 =
c
g1
Typical values for 0 are in the range of 1020 m2 1 2 . The relation between Einstein coecients
and the absorption coecient is in fact a bit tricky, notably because of the easy confusion between
spectral and spectrally integrated quantities19 .
As a nal remark, you should realize that all we treat here are approximations. Thus, the selection
rules work well in rst order, but if you look really in detail it turns out that they can be violated.
Typically, however, it is good practice to assume that the selection rules hold, and to start thinking of
forbidden transitions only if you really end up nowhere with the allowed ones.
Example: the transparent air
You may by now wonder why we see things around us at all. The light we need for vision must
pass through air, nitrogen and oxygen molecules with a total number density of about 2.5 1025
molecules/m3 . Why is this light not absorbed? Look at the potential energy curves of nitrogen and
oxygen (chapter 11.1.1, gures 11.3 and 11.4). Visible light corresponds to the wavelength range
of about 400800 nm, equivalent to the energy range of 1.553.10 eV. The atmosphere is at about
room temperature, say 300 K, equivalent to about 210 cm1 . The vibrational constants of oxygen
(1556 cm1 ) and nitrogen (2331 cm1 ) are much larger than that, so Boltzmann statistics tells us that
only the ground vibronic state will be appreciably populated. Homonuclear diatomics have no rovibrational transitions, so any transition must be to another electronic state. The appropriate selection
rules can then be used to nd out what the upper state might be:
Nitrogen: The ground state (X) has the term symbol 1 + , so the upper state must be 1 + or 1 u . The
g
u
diagram shows that there is in fact only one possibility: the b 1 + state. This transition would
u
require photons of at least 12 eV energy, corresponding to deep ultra-violet radiation of about
100 nm.
Oxygen: The ground state (X) is a 3 state, so the upper state must be 3 or 3 u . Inspection of the
g
u
potential energy curves shows that
1. the states at about visible energies above the ground state (that is, the a- and b-states) are
not accessible, because the transitions would violate the spin selection rule.
2. The lowest (bound) state that can be reached by an allowed transition is the B3 state,
u
which requires at least 6 eV photons ( 200 nm).
Thus, we nd that there are no allowed transitions for visible photons in the major constituents of
atmospheric air. Therefore, visible light traverses the atmosphere unimpeded, essentially unaware of
the immense number of molecules it passes through20 . The B-state of oxygen, by the way, can be
reached from the ground state by absorption of ultraviolet photons, 200 nm and below. In the upper
19
A very readable paper by R.C. Hilborn on this subject, Einstein coecients, cross sections, f values, dipole moments
and all that, can be found at http://arxiv.org/pdf/physics/0202029 .
20
This statement will be relaxed when we discuss non-resonant interactions, like Rayleigh scattering.
213
atmosphere this transition is responsible for ecient absorption of solar UV radiation. The absorption
leads to dissociation of oxygen molecules; the resulting atoms react with other O2 molecules to form
ozone. This process is responsible for the ozone layer in the stratosphere (Figure 11.13).
Figure 11.13: Processes related to the formation

and destruction of the ozone layer in the stratosphere. (Source: Wikipedia)
Absorption vs. emission

Distinction is made between absorption and emission spectra. An absorption spectrum reects the
wavelength-dependent absorption by a medium, observed if you send a light beam through it, like in
the basic experiment of gure 11.9. An emission spectrum corresponds to the wavelength dependence
of the light intensity emitted by the object of interest itself. Fundamentally, there is hardly any difference between the two types of spectra. Both concern light-induced transitions21 between quantum
states of the molecules involved, but in absorption a photon is lost, whereas in emission a photon is
produced. The selection rules are the same for both processes (a transition is a transition, irrespective
of the direction), and the only real dierence is that absorption can occur out of any initial state, but
emission must start in an excited state.
This dierence in initial state, albeit modest from a fundamental point of view, makes an important
practical dierence in the structure of absorption vs. emission spectra of the same molecule. One
obvious dierence is known as the Stokes shift of the emission spectrum: the emission spectrum on a
particular electronic transition is red-shifted relative to the absorption spectrum of the same transition.
Another obvious dierence is that the emission spectrum of a gas (that is produced upon exciting the
gas in some way or another) typically looks very dierent from the absorption spectrum of the same
gas under normal conditions, and often even falls in a completely dierent wavelength range. I will
briey discuss both dierences below.
The Stokes-shifted uorescence spectrum
Consider the situation of Figure 11.14. We look at the transition between vibrational states in two
dierent electronic states, that is, we look at vibronic transitions. Rotational ne structure is neglected,
for the sake of clarity. (How would it modify the spectrum?) We will assume that the temperature of
the gas is suciently low that only the ground vibrational state, v = 0, is populated. Absorption, then,
can start only in this ground state, and the nal state can be any vibrational state in the upper electronic
state (with transition strengths governed by the Franck-Condon factors.) In thermal equilibrium, there
21
What exactly induces emission is somewhat complicated, and in fact belongs to the realm of quantum optics. It boils
down to an interaction with the zero-point energy of the vacuum, but Im not sure that this knowledge will make you
particularly happy.
214
Chapter 11 N.J. Dam
Figure 11.14: Illustration of the origin of vibrational ne structure in an electronic transition, and of the Stokesshifted uorescence. Left: absorption out of the vibrational ground state; Right: emission out of the vibrational
ground state. The energy steps involved in emission are typically lower than those in absorption. (Remember
that the vertical direction corresponds to energy; the length of the arrows therefore is a measure for the photon
energy involved in each transition.)
will be no emission at all, since the upper electronic state is not populated. But now suppose we send in
a light beam tuned to the A(v = 0) X(v = 0) transition22 . Some molecules will absorb a photon out
of this beam, and become excited. Thermal equilibrium is then perturbed (in fact, you can not speak
of a temperature any more), and the excited state may relax back to the ground state by emission
of a photon. The possible nal states of this emission process are all vibrational states of the ground
electronic state, not only the v = 0 state. This is sketched in the gure. From the length of the arrows
(which correspond to the energy dierences between the connected states) you see immediately that
those involved in emission are shorter than the ones related to absorption: the uorescence is redshifted relative to the absorption. This same story holds if you excite higher vibrational states in the
upper electronic state, although the dierence becomes less pronounced. (Why?)
Collision-induced relaxation
If such an experiment is performed in practice on a gas sample at not too low pressure, the occurrence
of intermolecular collisions cannot be neglected. Such collisions may also induce transitions in a
molecule, and often they are quite ecient in doing so. Such transitions, however, are non-radiative.
Released heat now is dissipated into thermal motion of the collision partners: the gas heats up. We will
discuss this process (and a few related ones) more in detail later in the course, when we discuss the
laser-induced uorescence technique. For now, I only want to comment that even if higher vibrational
states in the upper electronic state are excited, collisions tend to take away the vibrational excitation
rather eciently, so that excited molecules end up in lower v states before they emit light. This serves
to restore the Stokes shift of the uorescence.
Emission spectra of excited gas
Under conditions of thermal equilibrium, the population is usually restricted to the electronic ground
state. Under non-equilibrium conditions, however, excited electronic states may also become populated, and then decay to lower electronic states by light emission. Such non-equilibrium conditions
22
Another convention: in this notation of a transition, the lower energy state is put at the right, the higher one at the left.
The direction of the arrow indicates whether you deal with emission or (in this case) absorption.
215
Figure 11.15: Spontaneous emission spectra of a slightly lean methane/air ame, with (bottom) and without
(top) plasma discharge. The assignment of each structure is indicated. The spectra are compilations of partspectra, with an overlap of about 10 nm.
can arise in various ways, for instance by chemistry (in ames), or by electrical discharges (in plasmas), or, more articially, by you (in doing Laser-Induced Fluorescence experiments)23 . Spectral
analysis of the spontaneous emission of a process can teach you a lot about the prevalent conditions,
but it must be remembered that it only provides information about the minority of species that is in an
excited state.
As an example, gure 11.15 shows the emission spectra of a methane/air ame, with and without
combustion assistance by means of a plasma discharge. Clearly, the presence of the plasma causes
a huge change in the emission spectrum. The dominant features in the normal ame are due to
the OH-radical (the A2 + X2 -band system around 300 nm, dominated by the (v , v ) = (0, 0)band) and the CH-radical (the A2 X2 -band system around 420 nm, dominated by the (v , v ) =
(0, 0)-band). Upon switching on the plasma, these bands remain visible, but in addition a strong and
extensive progression of bands in the so-called 2nd positive system of molecular nitrogen appears.
This name is traditional; it corresponds to the C 3 u (v ) B 3 g (v ) band system. This band system
involves a transition between two excited electronic states, and the de-excited molecules are therefore
still not back in the ground state. Thus, further spontaneous emission would still be possible.
23
Note that this should always be a concern to you: Am I really measuring the unperturbed ame, or do I produce
artifacts by the measurement itself?
216
Chapter 11 N.J. Dam
Question
In this particular case, decay to the electronic ground state is forbidden. Why? Can you nd another
electronic state to which N2 molecules in the B-state could decay by spontaneous emission? In what
wavelength range would you expect this spectrum? (Hint: look at gure 11.3.)
Spectroscopy software and literature
Spectroscopy is an ancient art. As a result, there is an enormous pile of literature available. Unfortunately, there is no recent compilation of spectroscopic constants, so that for any particular molecule a
literature search will have to be made in order to nd the required data.
Below I list a few general resources, and some recent free spectroscopy software. Both lists are
seriously incomplete, but merely reect what I have found useful over the past years.
Literature
1. G. Herzberg, Molecular Spectra and Molecular Structure, Vol. 14.

The classical work on spectroscopy, and in fact never surpassed. Volume 4 is a reference work
on spectroscopic constants of diatomic molecules; these data are also available from the NIST
Chemistry Webbook at http://webbook.nist.gov/chemistry/.
2. W. Gordy & R.L. Cook, Microwave Molecular Spectra.
A good second place, perhaps somewhat more readable than Herzberg.
3. R.W.B. Pearse & A.G. Gaydon, The identication of molecular spectra.
Classic tabulation of bandheads and persistent atomic lines, very useful when you need to identify the emitter of an unknown spectrum. Limited to small molecules.
4. W. Demtr der, Atoms, molecules and photons.
o
A modern introduction to molecular physics and spectroscopy. Emphasis on concepts. Buy the
rst edition, which is so full of errors that it enforces you to read carefully.
Software
LIFBASE Available from Jorge Luque, at http://www.sri.com/psd/lifbase/; you have to register, but
the software is free. It will calculate absorption spectra, excitation LIF spectra and emission
spectra of a few diatomic radicals, among which OH, NO and CH.
LifSim A web-based tool, available from http://www.lifsim.com/, that will simulate absorption, emission and LIF excitation spectra of O2 and NO.
PGOPHER Available from Colin Western, at http://pgopher.chm.bris.ac.uk/support.html. This is a
general purpose spectral simulation tool. You have to feed it spectroscopic constants, and it will
calculate absorption spectra in return. Very powerful, but it does require some experience in
spectroscopy.

Property
Frequency denition
Coherence
Spectral brightness
Pulse duration [s]
Beam divergence
Average power
Peak power
Cost of ownership
Cost of purchase
Ease of use
Perceived danger
217
Laser light
< GHz
high
high
1014
highly directional
several W
up to GW
high
high
low moderate
lethal
Classical light
> THz
low
low
103
extended source
up to kW
up to kW
low
low moderate
high
benecial
Table 11.4: Global comparison of properties of light sources. Note that the variation in light sources is enormous, so any comparison must be stated in terms of generalities . What I tried to do is to compare best
achievements, with an eye to applications in combustion diagnostics. (Specialized laser systems for particular
applications outside combustion may have much more stunning properties.)
11.2
Basic laser-diagnostic techniques
11.2.1
Equipment for laser diagnostics
One of the easily underestimated good things about laser diagnostics is that it requires you to spend
huge bags of money in purchasing fancy equipment, like lasers, CCD cameras and specialized optics.
Providing an overview of the market is a hopeless task, and developments are so fast that any overview
is bound to be obsolete as soon as it is printed.
In this section, I will merely provide a little bit of background in the basic characteristics of lasers
and detectors, as I have found them useful for combustion diagnostics. After all, the interest of this
course is in combustion, and the equipment is only used as a tool. I will not deal with optics at all.
Good introductions can be found in the catalogues of several suppliers!
LASER
The word LASER is in fact an acronym: Light Amplication by Stimulated Emission of Radiation.
Lasers are light sources with special, and sometimes very special, properties, if you compare them to
what I will call classical light sources, like res, lamps, or the sun, which are based on spontaneous
emission. A global comparison between laser light and classical light is provided in Table 11.4. All
lasers share the same basic ingredients (Figure 11.16), although the details vary, and the bells &
whistles of the gure will be decisive for the actual properties of a given laser system.
The operating principle of any laser is stimulated emission24 . Referring to gure 11.16 and gure 11.8, a very brief description of laser operation runs as follows. On switching on the energy source
(pump), molecules are excited and decay by spontaneous emission. This is a probabilistic event, and
by chance there will also be photons emitted along the optical axis of the cavity25 . These photons
24
There are other devices that produce radiation with similar properties like laser light, but by non-linear parametric
interactions. The possibly best-known example is the Optical Parametric Oscillator.
25
Note that rather than relying on chance, you can also provide the intitial photons externally, by another laser. This
process is called seeding.
218
Chapter 11 N.J. Dam
Figure 11.16: Basic ingredients of a laser. The active medium is pumped by some energy source (a ashlamp,
an electric discharge, an electric current, whatever) and emits light. After startup, those photons that run around
along the axis of the optical cavity are amplied (cloned, in fact). Losses through one of the mirrors provide
the output. The bells & whistles determine the detailed properties of the output.
will bounce back and forth between the mirrors, and are amplied each time they pass through the
medium. The amplication process imparts coherence onto the beam: stimulated emission produces
exact copies of the photons that do the stimulation. One of the cavity mirrors is deliberately < 100%
reective, and the leakage light through this mirror forms the output beam. Some brief notes on the
individual elements:
Active medium: This is the medium which emits light, and therefore is decisive for the global wavelength that will be produced. There are no fundamental restrictions on the nature of the active
medium, and it can be in gas phase, liquid or solid. Of course, there are practical issues to
consider. Notably, the uorescence strength should be competitive with the non-radiative losses
in the medium.
Pump: This is the primary energy source of the device. Its main purpose is to populate the upper level
of the laser transition in the active medium, so that stimulated emission dominates over induced
absorption. Examples include a broad-band ash lamp (Nd:YAG laser), electrical discharges
(many gas lasers), other lasers (dye laser), or simply an electrical current (diode lasers).
Bells & whistles: Determine the frequency bandwidth of the output, and whether it is pulsed or continuous (cw). A Q switch is a fast optical switch that renders the laser pulsed, with output
pulse durations in the order of a few nanoseconds. Etalons can be used to narrow down the
wavelength band that is amplied.
Cavity mirrors: These determine the beam axis, and the ne structure of the output wavelength band.
The latter arises because a cavity can only sustain wavelengths that exactly t into it, that is, the
cavity length should be an integral number of half-wavelengths. These wavelengths are called
cavity modes.
Characteristics of several specic laser systems, as they are often used for combustion diagnostics,
are collected in Table 11.5.
Frequency mixing
The response of a medium to incident light is linear in rst order, but for suciently strong irradiance
Diode laser
Nd:YAG laser
Ar-ion laser
Ti:Sapph laser
Dye laser
Excimer laser
Laser name
Active medium
(phase)
ArF (g)
KrF (g)
XeCl (g)
organic dye
in alcohol (l)
Ar+ ions
Ti impurities
in sapphire (s)
Nd impurities
in YAG crystal (s)
semiconductor
diode
ashlamp
or diode laser
electric
current
excimer or
Nd:YAG laser
discharge
cw laser
discharge
Pump
cw
pulsed (ns)
cw
pulsed (fs)
pulsed (ns)
Pulsed/cw
(typ. duration)
pulsed (ns)
4002000 nm
( 10 nm/diode)
1064 nm
Repetition
rate
193 nm
248 nm
308 nm
400800 nm
30 nm/dye
488, 514 nm
700-1000 nm
1500 mJ
10104 Hz
1100 mW
5W
1 J
Pulse energy/
Output power
100 mJ
300 mJ
100 mJ
10 mJ
80 MHz
10104 Hz
Output
wavelength
10100
Table 11.5: Properties of several laser systems often used for combustion diagnostics.
1 cm1
(seeded: 150 MHz)
1 MHz
few MHz
100 nm
0.1 cm1
1 cm1
Bandwidth

219
220
Chapter 11 N.J. Dam
the non-linear, higher-order response can no longer be neglected. A class of crystalline media exists
in which the second-order response is particularly strong, and these crystals are used routinely for
frequency mixing. The physical basis is not too complicated, and in fact rather similar to that of Raman
scattering. Now, however, the oscillating component in the polarizability is induced by the incident
light itself, rather than by molecular vibrations or rotations. The macroscopic, induced polarization in
the medium can be written as
P(t) =
(1) E(t) + (2) E 2 (t) + (3) E 3 (t) + . . . ,
(11.37)
in which 0 is the permittivity of free space (see table ?? and the (n) are the nth -order susceptibilities, tensors of rank n + 1. The (scalar) rst-order susceptibility is related to the (single molecule)
polarizability by
(1) = N ,
in which N is the local number density. Symmetry considerations require the even orders of the
susceptibility to be zero in many media (including all gases, liquids and symmetric crystals).
Let us now focus on those crystals for which (2)
0, and consider the 2nd -order polarization
induced in such a crystal by an incident monochromatic eld. Neglecting the vector nature of the
elds, we nd
2
1
P(2) (t) = 0 (2) (E0 cos t)2 = 2 0 (2) E0 (1 + cos 2t) .
(11.38)
Thus, the induced polarization, which acts as a source term for the electric eld, contains a dc term
as well as a term at twice the incident frequency (half the incident wavelength). This latter process is
called second harmonic generation. It can be very ecient, and theoretically even reach 100% conversion of the incident light. Slightly more general, the 2nd -order response to incident light containing
two distinct frequencies can be written as
P(2) (t) =
=
(2)
(E1 cos 1 t + E2 cos 2 t)2
(2)
1
2 0
2
2
E1 (1 + cos 21 t) + E2 (1 + cos 22 t)+
+E1 E2 cos(1 + 2 )t + E1 E2 cos(1 2 )t) . (11.39)

The induced polarization now contains a dc-term, 2nd harmonic terms of both input frequencies, as
well as a sum frequency term and a dierence frequency term. Not surprisingly, processes which
employ these two new components are called sum frequency generation and dierence frequency
generation. In the photon picture, all these processes correspond to the destruction of two incoming
photons and the creation of (at least)26 one new photon.
Although this is not evident from the above treatment, all these processes are highly directional,
and very sensitive to the proper alignment of the incoming polarization relative to the optical axis
of the non-linear crystal. The underlying concept is that of phase matching. In a non-linear crystal
the refractive index depends not only on wavelength (that is the case in all crystals), but also on
the polarization direction. For frequency mixing the orientation must be tuned in such a way that
the propagation speeds of all waves in the crystal are equal. In the photon picture, this condition
corresponds to conservation of linear momentum.
Frequency mixing is frequently used to extend the wavelength range of laser sources. A very
widespread application is frequency doubling of the output of a Nd:YAG laser (yielding 532 nm, bright
green), or its frequency tripling (actually frequency doubling followed by sum frequency generation of
26
At least??? Why is that?
221
the doubled light with the remaining fundamental beam; yielding 355 nm). With frequency doubling,
the eective tuning range of dye lasers can be extended down to about 200 nm. (Below this, air
becomes opaque. You should also realize that glass may no longer be transparent beyond the spectral
range that your eyes are sensitive to.) Dierence frequency generation extends the range to about
3 m in the infra-red.
Detectors
Over the past two decades or so, developments in CCD and CMOS technology have revolutionized
combustion diagnostics. Two-dimensional imaging of scattered light distributions has become the
standard, and adding a third dimension, in space or time, is now state-of-the-art. As with lasers,
the variety in commercially available detectors is way too large to review, so again I will restrict the
discussion to a few general features.
Broadly speaking, detectors can be classied into point-detectors and array detectors. The latter,
of course, can be seen as a collection of the former, but they make use of integrated circuit technology,
and therefore function as a single unit.
Point detectors integrate incident light over a certain photosensitive area and produce a single signal.
They do not oer spatial resolution by themselves. The photomultiplier tube (PMT) and the
photodiode are probably the best-known representatives of this class. Although the lack of
spatial information is a denitive disadvantage of these detectors (it is dicult to know what
you are looking at), they still oer superior sensitivity, speed, and spectral coverage.
Array detectors are one- or two-dimensional arrays of closely spaced photosensitive elements on a
single integrated circuit (chip). The individual detectors are, somewhat sloppy, called pixels,
picture elements. A typical detector may contain 1024 1024 pixels (1k1k) on an 18 mm
diameter chip. By using imaging optics in front of the detector, a 2-D region of interest can be
imaged onto the chip, which thus records photographs of the scene. In both types of camera
the pixels are small capacitors that collect charge in proportion to the incident light level. In
CCDs, each pixel is digitized sequentially, typically by only a single A/D convertor. In CMOS
chips, digitization (and possibly some other functions) is integrated in each pixel.
Two readable introductory papers have appeared in Photonics Spectra; they can be downloaded from
the Dalsa website. For scientic imaging, the current trade-o seems to be that CMOS cameras allow
larger frame speeds, but CCD cameras provide superior dynamic range.
Image intensication
Laser diagnostics of combustion often deals with low light levels of interest (e.g. uorescence of minor
species, or Raman scattering) against a high luminous background (the spontaneous emission of the
ame itself). An image intensier can help to counter both potential problems.
Image intensier technology is a whole topic in itself. Typically, the intensier concept resembles
that of gure 11.17. The whole device is inside a glass vacuum tube. Incident light illuminates a
photocathode, that is deposited on the inside of the front seal of the vacuum tube. The photocathode
converts light into electrons, which are accelerated towards a so-called multi-channel plate (MCP) by
means of a voltage dierence between the two components. The MCP is essentially a two-dimensional
array of small, slightly slanted tubes, the inside of which is coated by a thin metal layer. A high voltage
is present over the MCP, so that, essentially, electrons are forcefully pulled through the tubes. When
they hit the tube walls, a cloud of secondary electrons is emitted; this process repeats itself a few
times within each tube. What comes out is a cloud of electrons, that subsequently hits a phosphor
222
Chapter 11 N.J. Dam
Figure 11.17: Cross section of a generation-2 image intensier. (Source: Princeton Instruments, at the
Roper Scientic website.)
screen at the back end of the device. Here, the electrons are converted into photons again, and these
are imaged onto a CCD or CMOS chip either by a lens, or through a ber bundle (not shown in the
gure). Although the sensitivity of the whole device is reduced due to the conversion of photons into
electrons and back, the intensication can (largely) make up for that. At maximum gain, a single
detected photon may correspond to about 100 photoelectrons in the eventual image. Note that the
spatial information of the incident eld is retained, because intensication takes place within the
individual tubes of the MCP.
A major advantage of the image intensier is its capability of fast switching on and o, typically by
switching the voltage between photocathode and MCP. In combination with a pulsed laser, the actual
measurement may take place within a fraction of a microsecond. By keeping the image intensier
open during only that time, the background of natural ame luminosity can be suppressed by a factor
of 106 (for this example).
Spectral ltering
To glean full prot from the information present in scattered light, it is often a good idea to spectrally
resolve that light. Even if you are not particularly interested in the spectrum of the scattered light,
the information it contains may warn you for possible unexpected spectral interferences. If you are
intending a Rayleigh or Raman scattering measurement, for instance, then you would probably be
reassured by information that you are not accidentally including some unexpected uorescence signal
at the same time27 .
Filters, too, can broadly be classied in two classes.
Absorption lters transmit only the spectral range of interest, and absorb the rest. They used to be
rather unselective, but recent advances in coating technology now allow lters with a passband
of only a few nm wide to be produced for a few 100 euro. Filters are easy to use, especially
for imaging. Their major drawback in spectroscopic applications is that you can never be sure
about the actual emitter of light. After all, the only thing you know is that something emitted
light that was passed by the lter, but you have no information on the full spectral structure.
Dispersive lters in principle transmit all light, but in directions that vary with wavelength. Grating
spectrographs are probably the most often used example of such devices. In combination with
an array detector, such equipment allow one to record large parts of a spectrum simultaneously,
thus securing the maximum amount of information from the scattered light. The drawback is
that spectral resolution eats away one dimension: on a 2-D array, one axis contains spectral
information, so only one is left for spatial information.
27
This, in fact, was the case with the Raman images of gure 11.30. With an ArF excimer laser it is impossible to avoid
uorescence excitation in molecular oxygen. The alleged Raman image of nitrogen, therefore, is ok, but that of oxygen is
polluted by uorescence. Now think of this. How could you have told?
223
In a spectroscopic experiment, the need for spectral ltering cannot be overemphasized. If your
intention is to nd information on a particular chemical species, you must do every eort to ascertain
that this species is indeed what your signal is due to. Interference by other molecules is lurking around
every corner, and it is good practice to use opportunities of spectral selectivity wherever possible.
Safety issues
In real life circumstances you will unavoidably run into specic safety issues when you deal with lasers. The reason for
this can partly be found in table 11.5. Laser beams are highly
directional, so even a moderate power level can give rise to a
high irradiance. Pulsed lasers, moreover, concentrate power
in (very) short time intervals, so that peak powers may easily
end up in the 10100 MW range. Under these conditions,
even single laser pulses can incur considerable, and possibly
irreversible damage, especially to your eyes.
To my mind, working safely with potentially damaging
equipment should be a state of mind, rather than an ocial protocol that is forced upon you. Therefore, do not be
misled by the false feeling of safety that may be suggested
by external measures like laser goggles or interlocks. The
best guarantee to avoid accidents is a combination of knowledge and responsibility. Knowledge without responsibility
has produced atomic bombs, responsibility without knowledge has produced ARBO guidelines. The experimentalist
who knows what (s)he is working with and arranges the experiment in such a way that laser beams are conned to the
actual experiment, is no danger to colleagues and other innocent passers-by. If, at least, the combustion equipment under
study is safe itself...
11.2.2
Laser-induced uorescence
Arguably the most widely used technique for both the study of molecular spectra and the study of
combustion phenomena by means of spectroscopy is called laser-induced uorescence, or LIF. The
name is well-chosen: the technique is based on the detection of spontaneous light emission that follows (electronic) excitation by means of laser light (Figure 11.18). Thus, in its simplest form, LIF
is a time sequence of two one-photon processes28 : absorption of a laser photon on an allowed transition, followed by spontaneous photon emission out of the excited state. This somewhat resembles
the situation for Raman scattering of the next chapter (see gure 11.27, middle and right), but there is
one important dierence: the intermediate level is a real energy state in the case of LIF. This implies
that the molecule actually resides for a (short) while in this intermediate state, before it decays by
spontaneous emission or interaction with the environment29 .
In principle, any light source would do to induce the absorption step. Typically, a laser is used
because of its superior wavelength denition and spectral brightness. The spectral bandwidth of a laser
28
Variants exist, of course, e.g. 2-photon LIF, in which the rst step is a two-photon absorption.
Therefore, LIF is really a sequence of two 1st -order processes, whereas Raman scattering (next chapter), in which the
intermediate level is virtual, is a single 2nd -order process.
29
224
Chapter 11 N.J. Dam
often allows to resolve individual rovibronic transitions, a feature that would be very hard to achieve
using light sources based on spontaneous emission (i.e. lamps). By choosing the proper wavelength,
the transition that is induced in the molecule can be selected, often with high accuracy, and to the
exclusion of all other possible transitions. The subsequent emission step occurs spontaneously, and
the experimentalist has no control over it. Typically, there are many possibilities for the transition on
which spontaneous emission can take place, and it is not known beforehand which one any particular
molecule will decide to take. Nevertheless, the probabilities of occurrence of all transitions are
xed, so that the spontaneous uorescence spectrum of an ensemble of laser-excited molecules can be
predicted.
In practice, two kinds of spectra can be distinguished when using LIF. When the excitation laser
is scanned in frequency, and all ensuing uorescence is collected, a so-called excitation spectrum
is recorded. When, on the other hand, the excitation laser is xed on one particular transition but
the ensuing uorescence is spectrally resolved, we speak of an emission spectrum or uorescence
spectrum. Both kinds produce molecule-specic information.
The power of LIF as an experimental technique derives from two characteristic properties:
1. Signals are often strong, because both steps, absorption and emission, involve allowed transitions.
2. In gas mixtures (e.g. ames) it typically becomes easy to single out one particular chemical
species, because both the absorption and the emission spectra are characteristic for each component.
However, to quote a famous contemporary football philosopher: Efery advantuch hes its disadvantuch, and so it is for LIF:
3. The molecule remains for a nite time in its intermediate state, and during this time it is susceptible to interactions with the environment.
This disadvantage complicates the (quantitative) interpretation of LIF considerably, as we will see
below.
Figure 11.18: The principal steps in laser-induced uorescence. A laser excites molecules on one particular
transition, that is determined by the wavelength of the incident light. Subsequently, the excited molecules
spontaneously uoresce, for which typically many possibilities exist. Right: reduced model scheme. Note that
the vertical direction represents (internal) energy, the horizontal axis has no meaning.
225
The basic model system for LIF

In this section we will consider a model description of the signal that is typically measured in a LIF
experiment, that is, of the ensuing uorescence intensity. We will proceed in steps, adding complexity
(and realism) with each step.
As a basis, we just consider the three levels of the reduced scheme of Figure 11.18. We describe
the system in terms of the Einstein coecients of chapter 11.1.2. Rate equations for the level populations are as follows:
N1 = B13 ()N1 + B31 ()N3 + A31 N3
N3 = +B13 ()N1 B31 ()N3 A31 N3 A32 N3
N2 = +A32 N3
(11.40)
In these rate equations we have assumed that the incident radiation does not excite the transition
2 3, and that the uorescence is so weak that we need not be concerned with re-absorption. (There
is no loss channel for level 2.) We will assume that initially (that is, before we switch on the laser at
t = 0) only level 1 is populated:
N1 (t = 0) = N0
and
N2 (t = 0) = N3 (t = 0) = 0 ,
and that all levels are non-degenerate, so that B13 = B31 (see Eq. 11.17). Note that this model system
is a closed system, so that the total population N1 + N2 + N3 = N0 is conserved.
The spontaneous photons emitted on the transition 3 2 are detected. Obviously, the amount of
photons emitted up to a time t is equal to the population of level 2 at that time, N2 (t). We say that
level 2 is a sink for the population. The photon emission rate is equal to the rate of increase of N2 , and
thus proportional to N3 (t). We will now consider only the situation that N2
N0 , that is, short times
and/or low levels of incident light. In that case we can neglect the population loss channel to level 2,
and solve the rate equations for the other levels in steady state (that is, all time derivatives are zero):
B31 + A31
N0
B13 + B31 + A31
B13
ss
N3 =
N0
B13 + B31 + A31
ss
N1 =
(11.41)
(11.42)
For the signal photon rate, which is, apart from detection eciency issues, equal to the spontaneous
photon emission rate on the transition 3 2, we nd:
S = A32 N3 =
B13
A32 N0 ,
B13 + B31 + A31
(11.43)
in which the constant denotes detection eciency. Simple as this model may be, it already allows
us to draw a few important conclusions:
1. The signal is proportional to the initial population in level 1, that is, with laser-induced uorescence you essentially probe the initial level population.
2. For low excitation levels (low laser power), that is, if B13
with incident light irradiance:
A32 B31
N0 .
lim S =
0
A31
A31 , the signal scales linearly
226
Chapter 11 N.J. Dam
3. For high excitation levels, that is B13

dent of the actual laser intensity:
A31 the signal levels o to a constant value, indepenlim S = 1 A32 N0

2
This leveling o of the signal level with increasing laser power is called saturation. It arises
because also the level populations become equal: N1 = N3 = 1 N0 if : stimulated
2
emission and stimulated absorption balance each other. Note the practical implication: you
cannot increase signal level indenitely by just increasing the incident laser power.
4. Signal levels can be increased by selecting a transition with large A32 .
Improvements of the basic model
Several assumptions have been made for the basic model, that are typically not really valid in a practical situation. Many of these can be accounted for by the inclusion of additional terms in the rate
equations. Depending on your personal philosophy of life, these can be looked upon as improvements
or complications. Good to realize from the start, in any case, is that the additions typically do not
change the qualitative picture sketched above, but only the quantitative relation between signal and
actually present density.
Laser light is narrow-band, not broad-band
The rate equation description assumes incident light that is suciently broad to encompass the whole
linewidth of the transition of interest. Laser light can easily be narrower than that. If that is the case,
there are two limiting situations:
1. Rate equations may not be used any more. This is the case for narrow-bandwidth irradiation of
a system in which also the interaction with the environment (discussed further below) is weak.
We will not consider this case, even though it is of some practical relevance for non-linear
combustion diagnostics. When rate equations cannot be used, a more fundamental approach
involving the density matrix formalism must be used. This wildly exceeds the scope of this
course.
2. Rate equations can still be used, but the excitation eciency is reduced. This reduction comes
about because only a small part of the molecules is in resonance with the laser light (Figure 11.19), and can be accounted for by means of an overlap integral,
+
g(L , 0 ) =
Llaser (, L )Fmol (, ) d ,
(11.44)
which multiplies the energy density. This integral describes the weighted overlap of the frequency prole of the laser (prole L, nominal frequency L ) and the molecular transition (prole F, which is typically Gaussian, Lorentzian or Voigt, center frequency 0 ).
Steady state may not be reached
This is certainly the case immediately after switching on the laser. In practice, it is not usually a
concern, but merely contributes a scaling factor to the uorescence yield.
A case of great practical interest is that of LIF experiments with pulsed lasers. The pulse duration
of these lasers is typically about 5 ns, much shorter than typical natural lifetimes of excited states,
227
1.0
Lineshape
Line shape function (arb. units)
Laser profile
0.8
Figure 11.19: Example of a laser line (red) that

is clearly narrower than the molecular transition
(black). Drawing for the case that L = 0
400 MHz.
0.6
0.4
0.2
0.0
-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
(GHz)
which are 10s to 100s of nanoseconds. When such a laser switches on, it starts to pump population
out of level 1 into level 3, but long before steady state is reached it switches o again. In that case,
we can neglect the spontaneous decay of level 3 during the laser-on period, and let the population N3
decay spontaneously afterwards, detecting as many emitted photons as possible.
If we solve the rate equations of Eq. 11.40 under the condition of negligible spontaneous emission,
we nd
1
N1 (t) = 2 N0 (1 + exp{2B13 t})
N3 (t) = 1 N0 (1 exp{2B13 t})

2
(11.45)
For small durations of the excitation laser pulse, p , and assuming that we stay away from saturation,
we can make a series expansion of the exponent, to nd
N3 ( p ) B13 N0 p .
(11.46)
That is, the upper level population after laser excitation is proportional to the initial level population,
which itself, in this case, is equal to the total number density. Let us call this after-pulse population
(+)
N3 .
After excitation, the rate equation for the upper level 3 is simple:
N3 = A31 N3 A32 N3 .
(11.47)
So, if we start counting time at the moment the laser pulse is switched o, we nd
(+)
N3 (t) = N3 exp{(A31 + A32 )t} .
(11.48)
Only the photons on the transition 3 2 are detected. The emission rate on this transition is A32 N3 (t),
and so the total number of emitted photons, our signal S , is
S =
A32 N3 (t) dt =
t=0
A32
A32
N (+)
B13 p N0 .
A31 + A32 3
A31 + A32
(11.49)
Thus, the conclusions that we have drawn for the steady-state situation remain valid for this shortpulse experiment, too.
228
Chapter 11 N.J. Dam
Who tells the molecule to stop absorbing after one photon?

No one does. It is not impossible that an excited molecule absorbs another photon during the time
that the laser is on30 . The molecule will then reach a still higher excited state, and typically this state
does not give rise to detectable uorescence. In our model, we can incorporate this by means of a
bath of highly excited states, as in Figure 11.20. For simplicity, we will assume that once a molecule
arrives in this bath, it is lost31 .
Figure 11.20: Model system when absorption of a second photon is possible, bringing the molecule into a
highly excited state.
Continuing with the pulsed experiment of the previous section, the rate equations for levels 1 and 3
during the laser pulse now become
N1 = B13 ()N1 + B31 ()N3
N3 = +B13 ()N1 B31 ()N3 B ()N3 ,
(11.50)
in which B denotes the Einstein coecient for absorption into the highly excited states. Obviously,
the additional term adds to the stimulated emission term. The equations can be solved analogously to
eqs. 11.45, but the solution is immediately much more complicated:
N1 (t) = N+ exp{+ t} + N exp{ t}
1
N3 (t) =
(B13 + )N+ exp{+ t} + (B13 )N exp{ t}
B31
(11.51)
with boundary conditions N1 (0) = N and N3 (0) = 0, from which the constants N+ and N can be
derived. The decay constants are given by
1
= 2 (B13 + B31 + B )
1
2
(B13 + B31 + B )2 4B13 B
and the -dependence of the energy density has been suppressed. You might want to verify that in
the limit of B 0 the solutions 11.45 are correctly reproduced.
30
Not to be confused with 2-photon absorption, in which two photons are absorbed simultaneously in a single transition.
What I mean here is the sequential absorption of two photons, each one on an allowed transition.
31
A short comment may be in order here. Typically, the level density increases with increasing internal energy. Therefore,
it is likely that several states can be reached in the second absorption step, and this is the reason for introducing a bath rather
than a single state. Also, the number of perturbations increases strongly with increasing excitation energy (they include
processes like ionization and dissociation), and this explains my assumption that a molecule is lost for the measurement
once it ends up in the excited state bath.
229
Note that due to the loss channel in the excited states, the system is not closed any more, population
is not conserved, and in steady state N1 = N3 = 0. The important conclusions are that the nal
population in level 3 remains proportional to the initial lower state population N, but that a quantitative
interpretation now also requires knowledge of B . In practice, the latter is often not available, and for
this reason through-excitation should be avoided, when possible.
There are more levels than just these three
The three-level approximation is, of course, a coarse one, actual molecules possessing close to innitely many discrete energy levels. Thus, it would be better to consider an energy level scheme
like that in the left part of Figure 11.18 (but with still more levels). Let us now speak of groups of
levels: group 1 includes all potential initial levels, etc. To be specic, we might consider a situation
that occurs frequently: excitation starts in a rotational level of the vibronic ground state, so group 1
consists of all rotational levels of the X(v = 0) state. Excitation might be to a vibrational level in the
rst excited electronic state, so that group 3 would consist of all rotational levels in the A(v ) state.
The detected uorescence might arise from spontaneous decay back to any vibrational level in the
electronic ground state, so that group 2 would contain all rovibrational levels in the X-state32 .
First consider the multitude of levels in groups 1 and 3. Every level pair (level from group 1 plus
a level from group 3) has its own transition frequency, and it is questionable whether all of these are
suciently far away from the others to allow a laser to select them. This is for the experimentalist
to check beforehand: measure (preferred) or compute (2nd best) an excitation spectrum, and nd out
which transitions can be selectively excited. If possible, choose among those transitions for further
experiments. If simultaneous excitation of multiple lines cannot be avoided, you can try to take that
into account by adding appropriate levels to the level scheme. In general, this increases complexity
and uncertainty. I will further assume that isolated transitions can be used.
Because the uorescence step is spontaneous, and a probabilistic event, the multiple levels in
group 2 all need to be taken into account. They imply that even if only a single level in group 3 is
excited, the uorescence nevertheless will contain many wavelengths, corresponding to all allowed
transitions 3 2 j (subscript indicates individual levels in group 2). The intensity of each component
is determined by the size of the corresponding Einstein A-coecient, A32 j . For each component, there
are two possibilities: it either falls within the detector bandwidth, or outside of it33 . For the former,
these contribute to the uorescence signal; for the latter, they just represent a loss factor. In the model,
we can represent this by grouping all levels of group 2 into two bathes, see Figure 11.21. One bath
contains all levels for which the uorescence would be detected (labeled 2+ in the gure), the other
contains the rest, the uorescence to which escapes detection.
Assuming a pulsed laser experiment, the undetected bath simply represents another loss channel,
additional to the uorescence back to the original state 1. Analogous to Eq. 11.49 we now have
S =
A32+ N3 (t) dt =
t=0
A32+
A32+
(+)
N3
B13 p N0 ,
A31 + A32+ + A32
A31 + A32
(11.52)
in which now A32 is the total relaxation rate into level 2, A32 = A32+ + A32 .
The environment cannot be neglected
The environment contributes collisions. We have encountered them already in the lineshape-section
32
Note that there is overlap between groups 1 and 2; this is not a serious complication in practice.
If you are looking for trouble, you can also try to incorporate the wavelength dependence of the particular detector that
you are using.
33
230
Chapter 11 N.J. Dam
Figure 11.21: Extended level scheme, including selective excitation of one particular transition, and uorescence into two groups of levels. Fluorescence to bath 2+ is detected, whereas uorescence to bath 2 falls
outside of the detector bandwidth. The grey insert shows the spectrum as it might be observed, neglecting
rotation. Can you assign the transitions in the diagram on the left to the peaks in the spectrum?
of the previous chapter on spectroscopy (chapter 11.1.2), where we found that under atmospheric
conditions, molecules collide with each other at a rate of about one per nanosecond. For our present
purpose, only the inelastic collisions are of importance: collisions that cause the collision partners to
change state. It implies that in our model light is not the only agent any more that can change molecular states: collisions can do that, too. Unfortunately, the time scales of collision-induced population
transfer, and pulsed laser experiments and/or photon-mediated population transfer typically are all of
the same order of magnitude, so the occurrence of state-changing collisions cannot be neglected in a
realistic model for LIF.
The probability that a particular molecule changes state during an intermolecular collision potentially depends on many parameters: the chemical nature of the collision partner, the relative orientations and initial energies of the collision partners, the relative speed at which they collide, and possibly
some more. Obviously, trying to take all possibilities into account is a futile exercise, and inelastic
collisions are typically taken into account by rate equations, statements about the average behaviour.
Figure 11.22 shows a reasonably complete picture of all the disasters that intermolecular collisions
can bring about. The following processes are distinguished:
Rotational relaxation: the rotational quantum number changes (often by 1 per collision, as
indicated, but other changes are possible); the vibrational and electronic quantum numbers stay
the same.
Vibrational relaxation: the vibrational quantum number changes (often by 1 per collision,
as indicated, but other changes are possible); the rotational quantum number may change, the
electronic quantum number stays the same.
Electronic relaxation: The electronic quantum number changes, but the molecule remains electronically excited (see quenching), and the spin quantum number does not change (see ISC);
231
Figure 11.22: An overview of inelastic processes in a molecule. Energy is vertical. Energy levels (states) are
indicated by horizontal lines; their length has no meaning. Two inserts show a magnication. Lines with the
same colour belong to the same electronic state. Rotationless vibrational states are indicated by slightly fatter
lines; part of the rotational progression above these is indicated. Vertical thin dashed lines indicate that the
progression continues towards higher energies. Solid lines indicate radiative transitions, wavy lines indicate
collision-induced, non-radiative processes (and a few internal processes); all of these processes can actually
go in both directions, although only one is indicated. Collision-induced processes can connect any pair of states,
and obviously only a few examples are included. See legend and text for further explanation.
232
Chapter 11 N.J. Dam

the rotational and vibrational quantum numbers may change.
Quenching: Collision-induced, radiationless transitions back to the electronic ground state.
Internal vibrational relaxation IVR: Like quenching, but the nal state is a highly vibrationally
excited state at about the same energy as the original electronic state. The distinction with
quenching is not always made.
Intersystem crossing ISC: Like electronic relaxation, but with a change in spin quantum number.
Open channel losses: Any process, not necessarily collision-induced, that destroys the molecule
of interest. Examples are dissociation, ionization, or chemical reaction (electronically excited
states are typically highly reactive).
The radiative processes in the gure we already have encountered before, except for
Phosphorescence: Spontaneous emission on a forbidden transition.
Obviously, the combustion process that is our actual topic of study, itself contributes to some of the
above-mentioned processes. Combustion is nothing more than chemistry, the reshuing of electrons
during intermolecular collisions and some poor nuclei that are dragged along with them. The reactions
that keep the combustion going therefore contribute to the open channel losses.
Some important consequences of these collision-induced processes are the following.
Saturation is less easily reached, because rotational relaxation lls the depleted initial state of
the absorbing transition, and also removes population from the nal state.
After electronic relaxation, the molecule is still found in an electronically excited state, and
may uoresce. The uorescence then comes in an unexpected band, but may still be useful
for detection.
Quenching and IVR reduce uorescence, because they electronically deactivate molecules.
Intersystem crossing and phosphorescence typically are unimportant for small radicals; they
may play a role in the case of larger organic molecules.
Knowledge of rate constants for the collision-induced processes is scanty, and often the limiting
factor in the quantitative interpretation of the uorescence yield in terms of the number density
of the absorbing molecule.
In principle, it would be doable to set up a system of rate equations involving all relevant energy
levels of the molecule of interest, and all relaxation processes between them. The system would
be large and complicated, but essentially just a system of N coupled linear equations, and therefore
solvable, if only numerically. The exercise would be futile, however, since in practice not all required
rate constants are known.
In order to set up a simplied model system, there are a few important practical considerations.
1. Laser excitation is typically narrow-band, so that only a few, or in favourable cases just one,
excited state(s) are(is) directly excited.
2. The presence of rotational relaxation often precludes saturation, and tends to keep the initial
level population at its equilibrium value.
233
3. Fluorescence is typically detected in a wide wavelength range, encompassing many individual

transitions. As a result, many relaxation processes in the upper state need not have an eect on
the eective uorescence yield.
4. Collisional quenching tends to be the main uorescence loss channel.
Figure 11.23: Simplied level scheme for LaserInduced Fluorescence. Excitation out of level 1
with constant population, into bath 3, which contains all levels out of which the uorescence is detected. This upper level level bath experiences additional losses due to quenching and open channel
losses.
Taking these considerations into account, one arrives at the simplied level scheme of Figure 11.23.
Considering a pulsed-laser experiment again, the initial population of bath 3 is proportional to the original ground state population and the incident laser intensity. The upper state population then decays
according to
N3 (t) = (A32 + kQ + )N3

(11.53)
and the total uorescence yield, similar to Eq. 11.52, is given by
S =
A32 N3 (t) dt =
t=0
A32
B13 p N0 .
A32 + kQ +
(11.54)
In general, equations like the one above show that the eective uorescence yield is a ratio of decay
constants: the decay rate into the detected transition over the total decay rate. This ratio is called the
Stern-Vollmer factor. For the specic case above it reads
SVF =
A32
.
A32 + kQ +
The quenching rate constant, kQ , in this equation is an eective rate constant. In principle, a certain
collision partner is more or less ecient in bringing about collisional quenching, and we should in
fact write
( j)
kQ x j
kQ =
j
in which the summation runs over all chemical species present in the gas mixture with mole fractions x j . The individual quenching rate constants may vary appreciably with collision partner (e.g.
molecular nitrogen, N2 , is often relatively inecient, whereas water vapour, H2 O, tends to be a very
234
Chapter 11 N.J. Dam
Figure 11.24: Left: Potential energy curves of molecular oxygen, relevant for LIF in the Schumann-Runge
bands. Right: Spectrally resolved uorescence in the Schumann-Runge bands out of the directly excited level,
as a function of ambient pressure. In this case, the B(v = 6, J = 6) state is directly excited. See text for further
explanation. (Source: V. Sick et al., Chem. Phys. Lett. 249 (1996) 335340.)
ecient quencher), and they typically scale with the partial pressure of the component. Thus, kQ is in
principle a function of the detailed mixture composition. In ames, this often poses a limiting factor
to the accuracy of a LIF experiment, although the large number of participating species to some extent
provides a attening eect, rendering the quenching rate constant relatively unsensitive to the reaction
progress.
Some specic examples
Molecular oxygen, O2
The lowest-energy one-photon-allowed excitation of molecular oxygen out of the vibronic ground
state is into the B3 -state. All vibrational transitions in the B X electronic transition are collecu
tively known as the Schumann-Runge system, see Figure 11.24 (left). The potential well of the B-state
is shifted outwards relative to that of the ground state, and vertical excitations out of X(v = 0) into
low vibrational levels of the B-state therefore have a low probability. Another aspect of the B-state is
that, although it is a bound state by itself, its whole potential well lies above the dissociation limit of
the ground state34 . In the case of oxygen, internal couplings between several states cause the B-state
to be unstable: it dissociates quickly (time scales < 1 ns) into two ground state oxygen atoms. This
process, excitation into a state that is in principle bound, but nevertheless leads to dissociation to a
lower-lying dissociation limit, is called predissociation.
34
The ground (X) state is a bound state of two ground state oxygen atoms (this is what the symbols O(3 P) at the right
mean). The B-state is a bound state of one ground state oxygen atom and one excited state O-atom, O(1 D).
235
From a practical point of view, this gives rise to an interesting possibility for quantitative diagnostics. O2 can eciently be excited on the P- and R-branches of the B(v = 4, J ) X(v = 0, J =
1123) transitions by means of an ArF excimer laser (nominal 193 nm). The uorescence back
to (highly excited) vibrational levels in the ground electronic state occurs in the near-UV and visible
spectral range, and can eciently be detected by an ICCD camera or PMT. Chance has it that the
B(v = 4)-state is the most strongly predissociated, and most excited molecules dissociate rather than
decay by uorescence. In fact, under atmospheric conditions, the predissociation is so fast that it
dominates all other relaxation channels, even rotational including relaxation, which therefore can be
neglected. The Stern-Vollmer factor then reduces to
SVFO2 =
ABX
,
pre
in which pre is the rate constant for predissociation. This is an environment-independent parameter,
that can be found in the literature. Thus, quantication of the uorescence yield can proceed without
the requirement of detailed knowledge about the environment-dependent quenching rate constant. The
price to pay, however, is reduced signal strength: because pre
ABX , SVFO2
1.
The eect of the fast predissociation on the uorescence spectrum can be seen in Figure 11.24
(right). At low pressure (1 bar) the uorescence spectrum consists of only two peaks; these correspond to P- and R-transitions35 out of the directly excited rotational level in the B-state. From the fact
that only those two peaks are evident, it can be concluded that other rotational levels are not excited.
Apparently, then, there must be a process that removes population before it can decay rotationally:
this is the predissociation. With increasing pressure, the rate constants for collision-induced relaxation increase, and so the probability of rotational decay before dissociation increases as well. Thus,
uorescence on neighbouring transitions appears when the ambient pressure rises.
The hydroxyl radical, OH
In many types of ame, the hydroxyl radical is an abundant species. It is formed in the reaction
zone, and persists beyond that in quasi-equilibrium with water vapour. Because of its formation
history, it provides information on the location of the reaction zone, and therefore it is of interest to
experimentally quantify the OH-distribution in a ame.
Visualization of OH can proceed in two ways: either you use the spontaneous OH* chemiluminescence, or you use LIF. Chemiluminescence imaging has the advantage of simplicity (you dont need a
laser), but the disadvantages of providing only the OH*-distribution and of being a line-of-sight technique. Laser-induced uorescence has the advantages of complexity (you need a laser), info on OH in
the ground electronic state, and planar cross sections through the ame (no line-of-sight integration).
For obvious reasons (such as?) we will focus on laser-induced uorescence. OH is most conveniently excited in the 3064 A2 + X2 system36 . In contrast to the Schumann-Runge system of
molecular oxygen, the potential curves in this OH band system are very similar (Figure 11.25). As a
result, only v = 0-bands have appreciable transition strength: the band system is highly diagonal. It
might appear obvious to do both excitation and detection on the 0 0 band, but in doing so you might
run into problems with elastic scattering. In practice, therefore, one often excites in the 1 0 band
and detects diagonal uorescence (1 1 and 0 0, the latter out of the vibrational ground state that
is populated by vibrational relaxation). This is also the scheme that is indicated in the accompanying
gure.
35
36
Q-transitions are not allowed in this case, due to nuclear spin statistics.
For reasons unknown to me, this band system has never been baptized with a colloquial name.
236
Chapter 11 N.J. Dam
Figure 11.25: The hydroxyl radical, OH. Potential energy curves, an OH* chemiluminescence image of a
methane/air diusion ame (cylindrically symmetric around the vertical axis), and characteristic LIF excitation
spectrum around 300 nm.
Nitric oxide, NO
Nitric oxide is notorious as a pollutant produced during high-temperature combustion. Recently, it has
also been used as a photochemically produced tracer in molecular tagging velocimetry (MTV), kind of
a molecular variant of PIV. It can conveniently be detected in the so-called -system, A 2 + X 2 ,
the origin of which lies around 226 nm. For an illustration of the photophysics that may occur in LIF
experiments, another excitation pathway is more interesting: excitation in the -system, D 2 + (v =
0) X 2 (v = 1) by means of an ArF excimer laser at 193 nm. Following excitation of the D-state,
electronic and vibrational relaxation quickly leads to population in both the C- and the A-state, both
of which relax back to the electronic ground state by uorescence (and by quenching, of course, but
that we do not see); see Figure 11.26.
11.2.3
Rayleigh and Raman scattering
When energy conservation or selection rules prohibit the absorption of light on allowed transitions, it
still doesnt mean that light can traverse a medium unimpeded. In fact, in such cases we must look
at higher-order interactions. Two of these, Rayleigh scattering and Raman scattering, will briey be
discussed here. Characteristic for both is that the incident light does not match any allowed transition
in the molecules of interest.
The physical basis

Let us rst consider the word scattering. Scattering has a connotation of particle behaviour, and
indeed it is useful to look at scattering in the photon picture. I will call scattering an essentially
instantaneous process in which a photon that is incident on a molecule is destroyed, and a new photon
with dierent properties is created. If the new photon has the same energy as the incident one (but
237
Figure 11.26: Energy level scheme of NO, as far as relevant for LIF with detection in the -system. Rotational
levels are only shown in two inserts at the left. Below two experimental spectra, with assignment of the uorescence bands. The upper spectrum is observed upon excitation in the D 2 + (v = 0) X 2 (v = 1) band, the
lower spectrum is observed upon excitation in the A 2 + (v = 0) X 2 (v = 0) band.
a dierent direction of travel) we speak of Rayleigh scattering37 . If, on the other hand, the new
photon has a dierent energy than the incident one, we speak of Raman scattering. In both cases
the energy conservation law has to be obeyed. Thus, in the case of Rayleigh scattering no energy
is lost from the light, so also the scattering molecule must remain in the state it was in before the
scattering event. Raman scattering involves new photons with higher or lower energy (lower or higher
37
In fact, the scattering is not perfectly elastic, because of a (relatively small) Doppler shift invoked by the translational
motion of the scattering molecules. Most characteristic for Rayleigh scattering is that the scattering molecule does not
change state.
238
Chapter 11 N.J. Dam
wavelength) than the incident ones. The energy decit (positive or negative) must be accommodated
by the scattering molecule. As with uorescence, molecules can take up or release only discrete
amounts of energy, corresponding to the dierence between initial and nal states. Therefore, also the
dierence between incident and exiting photon energies must be discrete, matching that of the state
change of the scattering molecule. The big dierence with absorption or emission of light that we
discussed before, is that it is now the whole two-photon process (destruction & creation) that needs to
conserve energy. Rayleigh and the two versions of Raman scattering are illustrated in Figure 11.27.
Figure 11.27: Pictorial representation of scattering events, in which an incident photon of energy h is destroyed, and a new photon with the same or another energy comes out. Left: Rayleigh scattering, no energy
transfer; Middle: Stokes Raman scattering, energy transfer to the molecule; Right: anti-Stokes Raman scattering, energy transfer to the light. Solid horizontal lines indicate molecular energy levels, the dashed lines
indicate so-called virtual energy levels, that is, where an actual energy level would have to be to make it a
resonant process.
The intensity of the scattered light is usually expressed in terms of a scattering cross section, ,
kind of a geometrical surface area for removing light out of an incident plane wave. Alternatively, a
scattering coecient, R, can be used. In the linear regime, the intensity I s of the light scattered into
a particular direction (see gure 11.28) is linearly proportional to the incident light intensity I0 and
the illuminated volume38 , and is inversely proportional to the square of the distance L to the detector:
Is = R
I0 V
.
L2
(11.55)
The power incident on a small detector of surface area dA is

dPdet = I s dA ,
38
(11.56)
Or that part of the illuminated volume that is seen by the detector.
Figure 11.28: Parameters used in the description of scattering eciency. (After R. Boyd, Nonlinear optics, 3rd
ed., 2008.)
239
or, in terms of the solid angle d = dA/L2 that it subtends,

dPdet
= I s L2 = RV I0 .
d
(11.57)
Thus dened, the scattering coecient is a macroscopic parameter. The scattering cross section, ,
on the other hand is dened on a single molecule basis:
P1 = I0
(single molecule) ,
(11.58)
in which P1 denotes the total power scattered out of the incident beam by a single molecule. For a
particular scattering direction, we use the dierential scattering cross section, d/d:
dP1 d
=
I0 .
d
d
(11.59)
By comparison of eqs. 11.57 and 11.59 we can nd a relation between macroscopic and microscopic
parameters, if we assume that all molecules contribute independently to the total scattering signal:
R=
N d
d
=
,
V d
d
(11.60)
in which N is the total number of scattering molecules in the scattering volume V. The total power P s
radiated into a solid angle 4 is then
Ps =
d
d
V I0 .
(11.61)
The dierential scattering cross section varies with molecular species and with scattering mechanism.
For Rayleigh scattering, it can be written as
d
d
=
Rayl
42 (n 1)2 4
,
2 c4
(11.62)
whereas for Raman scattering there is no such simple expression. However that may be, it is clear that
both Rayleigh and Raman scattering eciencies are directly proportional to the local number density
(in the probe volume V). This makes the signals very easy to quantify in terms of local density, and
this is the power of techniques.
A simple, classical model for the Raman shift
Consider the following model for molecular scattering. We assume that the wavelength of the incident
light is such that single-photon allowed transitions are of no concern: the light is said to be oresonance. Treat the molecule as a positively charged core plus a negatively charged electron cloud
around it. The incident light, an oscillating electric eld, will set the electron cloud in motion relative
to the (much heavier) core; this is simply a forced harmonic oscillation. Suppose that, at the location
of the scattering molecule, the electric eld component of the incident light can be written as
E(t) = E0 exp{t} .
(11.63)
The induced polarization of a molecule hit by such a eld can be written as

P(t) = E(t) ,
(11.64)
240
Chapter 11 N.J. Dam
in which is called the polarizability. can be a constant, but if the molecule itself is, for instance,
vibrating, the polarizability will typically possess an oscillating component as well, say
(t) = 0 + 1 cos v t
(11.65)
in which 0 is the static component of the polarizability, 1 is the amplitude of the oscillating part,
and v is the angular frequency of the molecular vibration. Then
P(t) = (t)E(t) = 0 exp{t} Rayleigh
=
1
2 1 exp{(
1
2 1 exp{(
+ v )t} anti-Stokes Raman

v )t} Stokes Raman
(11.66)
Thus, the induced polarization, which acts as a source term for the scattered electric eld, now contains three frequency components: an elastic one, and two inelastic ones. The inelastic components
are symmetrically blue- and red-shifted relative to the elastic components, by an amount that corresponds exactly to the vibrational frequency of the molecule. We call the red-shifted component the
Stokes component, the blue-shifted one the anti-Stokes component39 , and the deviation from incident
frequency, v , is called the Raman shift. The above discussion is stated in terms of vibrations, but
it holds for any process that would produce an oscillating component into the polarizability, like, for
instance, rotation. Fact is, however, that Raman scattering for diagnostic purposes typically involves
molecular vibrations, and it will be the only form that we will consider here.
In the Maxwell equations, the induced polarization acts as a source term for the electric eld:
1 2 E
1 2 P
= 2 2 .
(11.67)
c2 t2
c t
From the frequency dependence of the induced polarization (eq. 11.66) it follows that this source term
scales with 2 (neglecting the relatively small Raman shift), so the scattered light intensity scales with
the 4th power of the incident frequency 40 . Also, insofar as is scalar, the scattered light retains the
polarization of the incident light.
2
Quantum renements
Two major shortcomings of the Placzek model for vibrational Raman scattering, as the above model is
called, are that i) it predicts that Raman scattering will not occur if the molecule doesnt vibrate (that
is, if it is in the vibrational ground state), and ii) that Stokes and anti-Stokes scattering are equally
strong. Both predictions disagree with observations. If you look at the energy level schemes included
in Figure 11.27, the reason for this disagreement is probably (hopefully) already evident.
In the quantum picture, molecules start out in a particular energy level, and their encounter with a
photon may result in them ending up in a dierent level. As usual, energy conservation requires that
any change in internal molecular energy is to be accommodated by the photons: if the molecule takes
up energy, the scattered photon will be red-shifted (Stokes-shifted) relative to the incident one, if the
molecule releases energy, the scattered photon will be blue-shifted (anti-Stokes). The amount of shift
(in terms of energy, h) is called the Raman shift, typically expressed in wavenumbers, cm1 . It exactly corresponds to the energy dierence between initial and nal levels of the scattering molecule.
Since this energy dierence is molecule-specic, the Raman shift is characteristic for a molecular
species. Table 11.6 lists the vibrational Raman shifts for a number of molecules relevant for combustion (diagnostics).
39
40
A terminology, by the way, that is not restricted to Raman scattering.

This explains the blue sky, as well as the red sunset, doesnt it?
241
Table 11.6: Characteristic vibrational Raman shifts (in cm1 ) for a few small molecules.
Molecule
O2
N2
H2
resonant
non-resonant
Raman shift
1556
2331
4160
elastic
Rayleigh
Molecule
H2 O
CO2
CO
inelastic
LIF
Raman
Raman shift
3657
1388
2145
Molecule
CH4
C2 H6
NH3
Property
species-specic
linear in power
linear in density
ease of quantication
strong signal
Raman shift
3000
3000
3300
Rayleigh
Raman
LIF
Table 11.7: Comparison of the three major linear light scattering techniques.
Question
Can you derive the values of the Raman shifts for the diatomic molecules in this table from the
molecular constants of Table 11.1? Which electronic state(s) would you need to consider?
Evidently, the intensity of a particular Raman line depends on the population of the initial state.
Thus, the intensity ratio of a Stokes/anti-Stokes pair is essentially determined by the population ratio of
the two states involved, eq. 11.14. This immediately suggests a major application of Raman scattering:
temperature measurement. After properly taking care of a possible wavelength dependence of the
detection sensitivity, the Stokes/anti-Stokes signal ratio is a direct measure for the Boltzmann fraction.
Raman scattering has its own selection rules, although these are not particularly strict. In practice,
v = 1 and J = 0 are dominant. Polyatomic molecules have multiple vibrational modes. Depending on the symmetry of the vibration, any particular mode may or may not contribute to the Raman
spectrum. The details go far beyond the scope of this course.
Practical issues and a few examples
Both Rayleigh and Raman scattering can be performed with any incident wavelength, which is a large
practical advantage over the use of LIF, in which a carefully chosen wavelength must be used that is
matched to a particular molecular absorption line. Rayleigh scattering, on the other hand, is elastic,
and doesnt carry species-specic information. Raman scattering is species-specic (via the Raman
shift), but has the serious disadvantage of low signal strength. The latter is perhaps not a problem in
principle, but it increases the risk of spectral interference by other processes with a larger scattering
eciency. Raman spectroscopy, therefore, requires extreme care in spectral ltering. Table 11.7
compares the three basic scattering processes.
Density eld measurements by Rayleigh scattering
Rayleigh scattering is often used for the visualization of density elds. The example of Figure 11.29
shows the idea. The experiment is conceptually straightforward. Using cylindrical optics, the output
of a laser (in this case an ArF excimer laser at 193.4 nm;41 ) was transformed into a thin light sheet
41
Remember that the Rayleigh scattering eciency scales with 1/4 !
242
Chapter 11 N.J. Dam
Figure 11.29: Rayleigh scattering images of an underexpanded nozzle ow. See text for details.
Figure 11.30: Raman scattering images of an underexpanded nozzle ow, perturbed by a metal rod (the red
disk in panels c and d). See text for details.
of about 22 mm wide and 0.5 mm thick. This light sheet passed immediately above the 5 mm diameter outlet of an underexpanded nozzle. This nozzle was operated with pressurized ambient air,
and produced a characteristic diamond pattern of consecutive high and low density regions along
the jet axis. The light scattered in a direction perpendicular to the light sheet was captured by an
ICCD camera. The laser is pulsed (< 20 ns pulse duration), so that the ow is essentially frozen in
its motion. Each pixel of a CCD camera can be considered to be a single detector, with a surface area
of about 20 m2 , onto which the light received from a specic probe volume illuminated by the laser
light sheet is projected. The total energy E p received by such a pixel follows from Eq. 11.61:
E p = P s exp =
d
d
V I0 exp ,
s
in which exp is the exposure time and is the light collection eciency of the detection system.
Several of the images (in a false colour scale) are reproduced in gure 11.29. All images show the
same region in space.
Panel a shows an image of the laser light sheet in ambient air, that is, without the nozzle ow.
243
Since the density is now uniform, this image shows the distribution of the laser intensity. Panel b
shows a vertical cross section through this image. Note that the upper part is not illuminated (the light
sheet was only 22 mm high) and there is a small intensity dip at around 15 mm (this is a particularity
of the laser used). When the jet is switched on (300 kPa backing pressure), the raw scattering images
look like that in panel c; you see structure at the location of the jet (goes up in the image, slightly
slanted), and the ambient air around it. The jet structure is highlighted in panel d, which shows
panel d =
panel c panel a
,
panel a
that is, panel d shows the dierence with ambient air density in units of ambient air density. Without
any further data treatment, panel d provides quantitative data on the density in the jet.
The major drawback of Rayleigh scattering is that it is very sensitive to light that is scattered o
any solid object in the neighbourhood of the probe volume. This is certainly a concern in conned
environments, and often precludes the use of Rayleigh scattering in practice42 . Figure 11.30 illustrates
the situation for the same jet experiment is described above. Now, however, a thin metal rod is put in
the jet, within the eld of view of the ICCD camera (panel a). The small rod scatters light so eciently
(relative to the air) that the ICCD camera is completely overexposed; panel b shows an image, rubbish.
Interestingly, much better results can be obtained in this case by using Raman scattering.
You should realize that Rayleigh and Raman scattering always occur simultaneously. Because the
former is so much more ecient than the latter, in a Rayleigh scattering experiment it is usually safe
to neglect the contribution of Raman scattering, even when no special measures have been taken to
suppress the Raman components. However, the Raman scattered light is shifted in frequency relative
to the Rayleigh scattered light. If a lter is used to selectively block the elastically scattered light (that
is, Rayleigh scattered light plus all the light scattered o the metal rod), only the Raman-scattered light
will reach the camera. This contains the same information (and a bit more) as the Rayleigh-scattered
light. Panels c and d of Figure 11.30 illustrate the above. For these images, the jet was operated on
pure N2 rather than air, at a stagnation pressure of 700 kPa, and the metal rod was inserted just below
the Mach disk that would form in an unobstructed jet under these conditions. In panel c the lter (a
spectrograph, in fact) selectively passed only the light that was Raman-scattered by O2 , in panel d the
lter was tuned to light Raman-scattered by N2 . Since the jet was operated on N2 , panel c is largely
empty in the jet area; O2 is only present in the surrounding ambient air, and apparently in the region
downstream of the metal rod. The nitrogen image of panel d, on the other hand, shows a distinct
density maximum just upstream of the metal rod, and a continuation of the diamond pattern in the
wake ow on both sides of it.
Example: Natural gas analysis by Raman scattering
Since Raman scattering is a non-resonant process, every component in a gas mixture scatters incident
light of any wavelength with its own characteristic Raman shift. This, at least in principle, allows
instantaneous gas analysis, without the need for multiple incident wavelengths (as would be the case
with LIF).
The experiment below shows how this works, for the example of natural gas analysis. Natural
gas typically consists of several components. The aim of the experiment was to nd out how accurate
each component in a certied synthetic natural gas mixture (Table 11.8) could be retrieved by Raman
scattering. The setup used is sketched in Figure 11.31. The light of a pulsed, frequency-doubled
Nd:YAG laser (532 nm) was used for irradiation, the gas under study was kept in a glass cell with
42
The recently developed technique of Filtered Rayleigh Scattering can overcome this problem, but it is a much more
complicated endeavor.
244
Chapter 11 N.J. Dam
Figure 11.31: Experimental setup for Raman analysis of natural gas.
Gas
Nitrogen
Carbon dioxide
Methane
Ethane
Propane
n-Butane
Helium
Formula
N2
CO2
CH4
C2 H6
C3 H8
C4 H10
He
CRB532 [nm]
607.0
571.0, 574.3
629.7, 633.8
561.5, 631.2
557.5
556.4
n.a.
% certied
3.877(15)
5.014(19)
87.70(22)
2.004(7)
0.7030(32)
0.1994(10)
0.5001(25)
% experimental
4.0(4)
4.2(4)
87.1(4)
n.d.
n.d.
n.d.
n.d.
Table 11.8: Gas components in the certied natural gas mixture, their characteristic Raman bands (CRB) for
excitation at 532 nm, and their concentrations according to the certication and the Raman experiment. n.a. =
not applicable (monatomic gases do not have vibrational Raman bands); n.d. = not detected.
Brewster windows (to avoid reections of the incident laser beam), and the scattered light was detected
by an intensied CCD array mounted behind a grating spectrograph. The grating available for these
experiments (1200 grooves/mm) was a bit inconvenient, in that only about 40 nm at a time could
be seen by the CCD array. The whole spectral range of interest therefore needed to be covered in
four consecutive recordings43 . Figure 11.32 shows the Raman spectra of the individual components
and of the synthetic gas mixture, all under nominally identical conditions of STP. The area below each
spectral peak has been tted to a model line shape, and, in line with Eq. 11.61, taken to be proportional
to the local density. By comparing corresponding peak areas in the syngas spectrum with that of the
pure component at STP, the relative density of each component was derived.
This example demonstrates several characteristic features of Raman scattering. The incident
wavelength is irrelevant, and individual components emit characteristic Raman bands. Similar components (like the hydrocarbons), however, also possess Raman bands in the same spectral region, so
that spectral interference may become an issue. Raman scattering is a weak process, given the rather
poor signal-to-noise ratio of the experimental spectra. (Note that the Rayleigh peak around 532 nm
appears huge, even though it is saturated.) Finally, however, even the very non-sophisticated data
treatment already results in accurate values for the concentrations.
43
By using a 300 grooves/mm grating this whole spectral range could have been recorded simultaneously, be it at the
expense of spectral resolution.
245
Figure 11.32: Raman spectra of the individual components and of the synthetic natural gas mixture. Vertical
lines indicate characteristic Raman bands. The fast rise to saturation at the left is due to the nearby incident
light wavelength, that is not fully rejected by the spectrograph. Each spectrum consists of four part-spectra, and
no attempt has been made to avoid breaks in the vertical axis.
246
Chapter 11 N.J. Dam
Chapter 12
Recent developments in laser diagnostics

M. Ald n
e
This chapter contains the rst sections of a paper presented by M. Ald n at the 32nd Int. Combustion
e
Symposium in Beijing, China. For the last sections, the reader is referred to the full article in Proc.
Combust. Inst. 32.
12.1
Introduction and background
Despite its impressive age, combustion is still of crucial importance in many pivotal human activities,
e.g. heating, transportation, traveling and waste handling, and it will remain so for years to come.
The economies of many countries rely strongly on combustion for propulsion and power generation
in industries producing cars, trucks and related products. Research which will allow us to improve the
eciency and cleanliness of engines, gas turbines, furnaces, to optimize chemical processes, decrease
the damage of res, etc., will contribute signicantly to competitiveness, sustainability and thus to the
environment, economy and employment, and to the quality of life of people throughout the world.
In order to optimize and control combustion processes it is of crucial importance to have techniques which are able to measure critical parameters, such as temperature, species concentration,
velocity, and particle characteristics. The challenges in this environment can be substantial; high
temperature, high pressure, highly turbulent gas ows, and a variety of fuels, often involving phase
transitions and thus multi-phase environments. Over the last few decades, perhaps the most important development for a deepened understanding of combustion processes, in terms of experimental
achievements, has been the introduction of various laser diagnostic techniques.
The main advantages with these techniques are that they are non-intrusive, i.e. measurements
can be made without introducing a physical probe that may inuence measurements by altering the
chemistry and/or uid dynamics, and that measurements can be made with very high spatial ( 10
50 m) and temporal (< 10 ns) resolution. Thanks to these features, laser techniques have been
successfully applied in characterization of combustion/ow phenomena. Since the techniques covered
in this review are based on high power lasers, which enable single shot two-dimensional visualization,
the general drawbacks with these techniques are the need for rather sophisticated equipment and a
relatively high cost. It should also be mentioned that the techniques described rely on short high
peak power lasers where eye safety is an issue. Thus all measurements described require proper and
adequate protection against any potential eye damage.
The main scientic directions in the area of development and application of laser diagnostic techniques for combustion studies may be divided in various categories according to:
247
248
Chapter 12 M. Ald n
e
1. Development of new diagnostic techniques as well as fundamental studies of both new and
established techniques to increase their accuracy and precision. This category may involve
rather fundamental aspects of atomic, molecular, and laser physics, and non-linear optics, and
may not necessarily aim for a deepened understanding of combustion per se but rather the
measurement techniques.
2. Applications of the more mature techniques for measurements of relevant parameters in laboratory ames for studies of various combustion phenomena, e.g. turbulent combustion, soot
formation, and NOx formation. These experiments often take place in close collaboration with
experts from other areas, such as chemical kinetics, uid mechanics, etc.
3. Applications of mature techniques for characterization, optimization and control of industrial
processes, i.e. in situ measurements in engines, gas turbines or furnaces.
The separations between these areas are not distinct and in reality large overlaps exist, relying on a
close interaction of people and a ow of expertise and equipment between the dierent tasks.
The activities directed towards development and application of laser techniques for combustion
studies has a long history and one of the rst, if not the very rst, conference in the eld was arranged
in Schenectady in 1974 [1]. This was the start of a strong development, mainly in the rst category
above, involving the use of more powerful lasers and multidimensional detectors, new non-linear optical techniques, e.g. [2-5] and multi-point imaging [6-8]. Already during the eighties there were thus
invited talks at the Combustion Institute, e.g. [9-11]. During the following decades many applications
for studies of combustion phenomena appeared, and now the focus has moved to category two and
three as shown in Invited talks at the Combustion Institute during the nineties, and the rst decade
of this century [12-14]. Today, the application of laser diagnostics in combustion processes is wide
spread. Aside from university activities, there are also governmental laboratories focusing on these
areas in both the US and Europe, and around the globe.
Various techniques have been developed and used for measurements of more or less all parameters
of importance for characterization of combustion phenomena, i.e. species concentrations; molecules,
radicals and atoms, temperatures; rotational, vibrational, translational and electron- temperatures,
velocities, surface measurements and particle characterizations; sizes, number densities and volume
fractions. The techniques which have been used can be split into spontaneous techniques based on
linear optics, e.g. laser-induced uorescence, Mie, Rayleigh and Raman scattering, where the signal
is emitted in more or less all directions, i.e. 4 solid angle, and coherent techniques based on nonlinear optics, e.g. Coherent anti-Stokes Raman Scattering, CARS, Polarization Spectroscopy, PS, and
Degenerate Four-Wave Mixing, DFWM, where the signal is generated as a laser-like beam. For an
overview of dierent techniques and their application in various combustion applications, in addition
to the references above, we also refer to Refs. [15,16].
In the present paper we will present some activities representing categories 1-3 above. Due to
space limitations, the main contributions will come from the authors laboratory, but key references
and contributions from other similar activities worldwide will also be given. The focus will be on
applications of laser-induced uorescence, including the use of a high-power, single-mode tunable
Alexandrite laser, two-photon LIF and high speed visualization, the development of a technique for
spatially resolved absorption-based measurements in the IR spectral region, polarization spectroscopy,
and the use of thermographic phosphors for temperature measurements, mainly on surfaces but also
in droplets and gas ows. Special emphasis will also be put on special challenges requiring new
techniques, e.g. measurements in combustion environments with very limited optical access, requiring
single-ended experiments, and measurements in an environments where multiple scattering may limit
249
the use of conventional techniques, e.g. in sprays. In the latter areas mainly two techniques will be
exemplied, namely picosecond LIDAR and Structured Laser Illumination Planar Imaging, SLIPI,
respectively.
12.2
Further developments and applications of planar laser-induced uorescence
Laser-induced uorescence, LIF, is a widely used optical measurement technique in combustion diagnostics with the strength of imaging capacity, high sensitivity, high spatial resolution, species specicity and applicability in harsh environments.
LIF is based on laser excitation of atoms or molecules from a lower energy state, most often the
ground state, to a higher excited state. Upon relaxation back to the lower state the atom/molecule reemits radiation. This radiation can be analyzed and information on both temperature and concentration
can be deduced. Since the excitation process is species dependent and the uorescence from the upper
state also occurs from, in the case of molecules, denite/discrete vibrational and rotational levels, the
selectivity of the LIF technique is considerable. This selectivity combined with high sensitivity makes
LIF a very powerful diagnostic tool. However, a major challenge with LIF is to convert the number
of photons emitted to the number of molecules of the species of interest. The main problem is to
account for de-excitation of the excited species through collisions of the excited species with other
molecules/atoms, i.e. quenching. Since the number of species de-excited due to quenching is typically
about three orders of magnitude larger than those giving rise to uorescence, this loss mechanism is
severe. Furthermore, since the quenching is dependent on colliding species, temperature and pressure,
it is a delicate task to convert the detected LIF signal into number densities, especially for data taken
in a turbulent environment.
Planar Laser-Induced Fluorescence (PLIF) has since the early 1980s become one of the most
powerful tools for studies of turbulent combustion phenomena, both for fundamental and applied
research. Enormous research eorts have been made in developing dierent PLIF techniques for
quantitative and precise measurements in various practical combustion processes. For overviews of
PLIF applications in combustion research, see e.g. [15, 16, 17, 18].
In the current presentation, rather than aiming at reviewing various applications, some further
developments of the PLIF technique, involving their applications, are given below.
12.2.1
Use of an Alexandrite laser
Clearly, single-shot PLIF is one of the most important tools for the investigation of turbulent combustion, revealing the instantaneous ame structure to increase the detailed understanding of turbulent
combustion and to visualize the combustion/ow ne structure in applied combustion apparatus, e.g.
engines and gas turbines. For this purpose sucient laser energy distributed in a plane is a necessity.
So far in the history of PLIF almost all measurements have been performed using either a Nd:YAG or
an Excimer laser, often in combination with a dye laser. However, one drawback with these systems is
that the pulse duration is on the order of <15 ns, which may result in excessive laser intensity, driving the molecular/atomic transition to saturation. In this case the uorescence emission is limited and
can not respond to an increase in laser power, which thus sets a limit on how strong the PLIF signal
would be. To further increase the signal intensity, and to avoid saturation, it could be advantageous
to increase the laser pulse length and/or the laser linewidth, in the latter case to excite more than one
rotational line.
In order improve the signal-to-noise ratio in PLIF measurements the use of high power excitation
laser pulses with longer duration and a broader linewidth to increase the PLIF signal intensity have
5 cm. Based on the laser specifications, a signal enhancement of about two orders of
magnitude was estimated as compared to conventional Nd:YAG pumped dye or OPO
asers, which enables the detection of CH down to sub ppm levels [20]. Shown in Figure
1 is a single shot PLIF image of CH in a highly turbulent jet flame when a laser sheet of
90 mm was used. 250
Chapter 12 M. Ald n
e
Figure 12.1: Photograph of a piloted jet

ame and a single-shot CH PLIF image
taken from the cross section through the
symmetric axes.
Figure 1. Photograph of a piloted jet flame and a
single-shot CH PLIF image taken from this purpose an Alexandrite laser system [19] has been utibeen investigated and demonstrated. For the cross
section through the symmetric axes.
lized which fullled all the properties mentioned above. The laser running at 10 Hz has a tuneable
wavelength covering 740 to 790 nm and a maximum pulse energy of 400 mJ. The laser can be operated either injection seeded for single-longitudinal mode or without seeding for multi-mode operation
The high signal-to-noisearatio achievable withcm1PLIFamakes duration of 100powerful
with linewidth of about 8 CH and pulse the technique a ns. Detailed investigations on how this
laser can be used for increased detectability of CH conditions where
ool for flame front structure investigations, especially for turbulentand HCO have been made. Both of these species
are frequently used as ame front markers and CH is an intermediate in NOx formation processes
dramatic changes in local equivalence ratio occur, resulting in a high dynamic range of
whereas HCO is used for spatial localization of maximum heat release, which makes them attractive
for single shot two dimensional imaging.
For CH PLIF detection, the transition X-B(0,0) was adopted for excitation with the multimode
9
laser frequency doubled to around 387 nm to cover several rotational lines in the molecular band
head. Following excitation, the uorescence from the B-X(0,1) and A-X transitions around 430 nm
were collected and detected with an ICCD camera. Around 70 mJ per pulse of the broadband 387
nm laser radiation was found to saturate the CH PLIF signal in an atmospheric pressure ame using a
laser sheet with a height of 5 cm. Based on the laser specications, a signal enhancement of about two
orders of magnitude was estimated as compared to conventional Nd:YAG pumped dye or OPO lasers,
which enables the detection of CH down to sub ppm levels [20]. Shown in Figure 12.1 is a single shot
PLIF image of CH in a highly turbulent jet ame when a laser sheet of 90 mm was used.
The high signal-to-noise ratio achievable with CH PLIF makes the technique a powerful tool
for ame front structure investigations, especially for turbulent conditions where dramatic changes
in local equivalence ratio occur, resulting in a high dynamic range of the CH concentration. This
technique has also been applied together with a conventional OH LIF technique for simultaneous
single-shot imaging of CH and OH radicals using PLIF to investigate partially premixed turbulent jet
ames. Various ames have been stabilized on a coaxial jet ame burner consisting of an outer and
an inner tube of diameter 22 and 2.2 mm, respectively [21]. From the outer tube a rich methane/air
mixture was supplied at a relatively low ow velocity, while a jet of pure air was introduced from the
inner one, resulting in a turbulent jet ame on top of a laminar pilot ame. The turbulence intensity
was controlled by varying the inner jet ow speed from 0 up to 120 m/s, corresponding to a maximal
Reynolds number of the inner jet airow of 13,200. The CH/OH PLIF imaging clearly revealed the
local structure of the studied ames. In the proximity of the burner, a two-layer reaction zone structure
was identied where an inner zone characterized by strong CH signals has a typical structure of rich
the first time enabled single shot HCO PLIF measurements in turbulent jet flames [22].
Shown in figure 2 is a series of single shot HCO PLIF images taken in a premixed
methane/oxygen flame burning on a 2 mm diameter welding torch burner.
251
Figure 2. HCO PLIF in a CH4/O2 welding torch flame: a) 100 single-shot
Figure 12.2: HCO PLIF in a CH4 /O2 welding torch ame: a) 100 single-shot averaged image; b)-e) consecuaveraged image; b)-e)
tively recorded single-shot images. consecutively recorded single-shot images.
premixed ames. An outer reaction zone characterized by strong OH signals has a typical structure
Two-photon LIF
of a diusion ame that oxidizes the intermediate fuels formed in the inner rich premixed ame. In
Other important species in combustion where PLIF closed surfaces have been sparse, are
the moderate-turbulence ow, the CH layers were very thinmeasurements extending over the entire
ame, whereas the OH layers were much thicker. In the high intensity-turbulence ame, the CH layer
for example light atoms; H, O, C, N and molecules like CO, H2O, and NH3, which have
remained thin until it vanished in the upper part of the ame, showing local extinction and reignition
their absorption lines located images have been utilized to i.e < 200 nm.. Due to the density
behaviour. The single-shot PLIFin the VUV spectral region, determine the ame surface strong
(FSD). In low and moderate turbulence intensity cases, the FSDs determined from CH and OH agreed
absorption by air and the difficulties generating tunable radiation with sufficient power in
with each other, while in the highly turbulent case a locally broken CH layer was observed, leading
to a signicant dierence in the FSD resultsone-photon via the OH and CH radicals. Furthermore,
this wavelengths regime, conventional determined LIF is not applicable. Instead, a twothe mean values and the corresponding standard deviations of CH and OH radicals were extracted
photon excitation information about the
to provide statisticalprocess can be used. ames, which can be used for validation of numerical
calculations. HCO is another important radical present in most hydrocarbon ames. The same idea as
described above for differs from single-photon LIF mainly PLIF measurements, which potentially
Two-photon LIF CH has also been used for improving HCO in the following respects:
can be used as a generic ame front marker. Using the 3rd harmonic of the Alexandrite laser, a 10
mJ UV beam at - Different selection rules 120 ns pulse duration and 10 cm1 laser linewidth.
258 nm was produced with
The application ofDoppler-free measurements are the rst time enabled single shot HCO PLIF
- such an excitation pulse has for possible
measurements in turbulent jet ames [22]. Shown in gure 12.2 is a series of single shot HCO PLIF
The temporal behavior ame excitation 2 mm diameter welding torch burner.
images taken in a-premixed methane/oxygen of theburning on apulse is important
- The LIF signal is proportional to the laser-intensity squared (ideally)

12.2.2
Two-photon LIF
The last point is, from an experimental point of view, very important. This characteristic
Other important species in combustion where PLIF measurements have been sparse, are for species
means that if a laser beam is focused in an environment with constant example
light atoms; H, O, C, N and molecules like CO, H2 O, and NH3 , which have their absorption lines
concentration, the two-photon LIF <200 will be to the strong absorption by air and the diflocated in the VUV spectral region, i.e signalnm. Due highest in the focal region of the laser
culties generating tunable radiation with sucient power in this wavelengths regime, conventional
beam, in contrast to single-photon LIF where the signal intensity is independent of
one-photon LIF is not applicable. Instead, a two-photon excitation process can be used. Two-photon
LIF diers from single-photon LIF mainly in the following respects:
Dierent selection rules
Doppler-free measurements are possible 11
The temporal behavior of the excitation pulse is important
252
Chapter 12 M. Ald n
e
The LIF signal is proportional to the laser-intensity squared (ideally)
The last point is, from an experimental point of view, very important. This characteristic means that
if a laser beam is focused in an environment with constant species concentration, the two-photon LIF
signal will be highest in the focal region of the laser beam, in contrast to single-photon LIF where the
signal intensity is independent of focusing since the increased laser intensity, I, is exactly balanced by
the decreased probe area, A, i.e.
Ione-photon I/A A = I
Itwo-photon (I/A)2 A = I 2 /A
Despite its low transition probability, two-photon LIF was rst demonstrated in the early eighties for
point-wise detection of some species e.g. O [23] and H [24]. Furthermore, PLIF measurements of
these species have been demonstrated [25], even for engine conditions, e.g. H2 O [26] and CO [27].
Carbon monoxide, CO, is an important species, not only because it is a major pollutant, but it can also
be used as an indicator of combustion eciency, since incomplete oxidation of hydrocarbons results
in increased concentration of CO in the burnt gas. Figure 12.3 shows single shot CO PLIF images
recorded in an SI engine at dierent crank angle degrees [27]. Compared with ame chemiluminescence images, which were also recorded, it was found that although a strong ame chemiluminescence
was observed at 30 CAD after TDC, the CO LIF had already decreased.
Clearly, quantication is an even more challenging issue for two-photon LIF, being a non-linear
technique, than for single-photon LIF. To achieve a quantitative interpretation of the two-photon LIF
signal, substantial eorts must be put into measuring important parameters such as the two-photon
absorption cross-sections, photo-ionization cross-sections, ac Stark shifts, collisional broadening and
shift, and species-specic and temperature-dependent quenching cross-sections. Furthermore, many
of the species requiring a two-photon excitation step are excited in the UV spectral region, e.g. H 205 nm, N - 211 nm, O - 226 nm, CO - 230 nm. To compensate for the weak cross section of the twophoton excitation, high peak power lasers are commonly applied, which consequently may introduce
strong interferences from fragments generated by photolysis, especially when species are to be probed
in a combustion environment. For example photochemical eects from vibrational hot CO2 [28] has
been investigated as a source of interference for two-photon LIF CO detection in atmospheric pressure
ames. Also C2 has been found to be a potential source of spectral interferences in detection of CO
[29]. For the engine results displayed in Figure 12.3, both the interference from C2 and photolysis of
hot CO2 was found to be negligible based on spectrally and spatially resolved measurements.
In order to address the problem with photo-production of species, and especially when detecting
atomic species, it has been proposed to use a picosecond laser for excitation [30], considering the
linear response of the photolysis production and nonlinear response of two-photon excitation. The
application of picosecond laser pulses thus dramatically increases the peak power to keep the twophoton excitation eciency but decreases the laser pulse energy to reduce the photolysis interference.
This technique has been applied for oxygen atom detection in ames providing interference free O
concentrations [31, 32]. Two-photon excitation of atomic hydrogen with a 205 nm laser beam was
also investigated with combined ns and picosecond laser pulses [33,34] and the photolysis interference
precursors have been investigated using the spatial distribution and concentration proles [34].
12.2.3
High speed visualization
Until the late nineties many laser diagnostic combustion applications were made using commercially
available Nd:YAG lasers, which were limited to repetition rates in the range 1050 Hz. Such temporal resolution excludes the possibility of following relevant turbulent phenomena in real-time. For
of hot CO2 was found to be negligible based on spectrally and spatially resolved
measurements. Recent developments in laser diagnostics
253
Figure 12.3: Single shot two-photon LIF

images of nascent CO from an SI engine
recorded at dierent crank angles.
Figure 3. Single shot two-photon LIF images of nascent CO

from an SI engine recorded at different crank angles.
example in engine applications the limitation in repetition rates implies that only one image can be
captured per cycle. Excimer lasers can be slightly faster especially when
In order to address the problem with photo-production of species, andbut still not sucient for temporally resolving
turbulent time scales. Cu+ lasers can reach repetition rates of tens of kHz, but the low pulse energies
detecting atomicmake them suitable only for Mie scattering applications. The excitationapproach to circumvent the
species, it has been proposed to use a picosecond laser for standard
lack linear response of the photolysis production and nonlinear response
[30], considering the of sucient temporal resolution, when following rapid events, e.g. in turbulent ames, is to measure probability density functions by capturing a suciently large amount of data to obtain adequate
of two-photon excitation. The application of picosecond laser pulses thus dramatically
statistics. In the case of studying IC engines, e.g. spray development or ame propagation, the only
increases the peak power to keep the data from dierent cycles at subsequent crank angles. The recorded data is then
solution is to capture two-photon excitation efficiency but decreases the
put together the photolysis interference. This If only average properties are
laser pulse energy to reduce in order to cover the entire event. technique has been applied of interest, the limitation
of one image per cycle is acceptable. However, for processes with low cycle-to-cycle stability, the
for oxygen atom detection in flames providing interference free O process with low stability is the spatial distribusingle image approach will suer. One example of a concentrations [31,
tion of combustion initiation and with a 205 Homogeneous was also
32]. Two-photon excitation of atomic hydrogenprogression innm laser beam Charge Compression Ignition (HCCI)
engines. Other and picosecond laser pulses [33,34] and the photolysis
investigated with combined ns examples of unstable processes are ignition and early ame development in directinjection spark-ignited engines (DISI) where it is essential that the stratied fuel charge is transported
to the vicinity of the spark plug in the correct quantities and that it arrives at the time of ignition. This
far from stable process shows a signicantly increased number of misres compared to conventional
SI engines.
13
High-speed laser diagnostics may allow for resolving turbulent processes and for single-cycleresolved engine measurements, in which single combustion events can be followed in time. Cycleto-cycle variations of complex events can thus be studied. As individual structures can be followed
in time, the large-scale development of e.g. fuel injection, ignition events, or ame propagation, can
be visualized. This is crucial when trying to identify the origins of instabilities, e.g. in the mixture
254
Chapter 12 M. Ald n
e
transport.
To realize the need for high speed visualization, essentially two approaches have been developed
during the last decade, and these are briey described below.
Nd:YAG Laser Cluster - High speed measurement
In order to perform high-speed diagnostics a suciently fast excitation source is needed. One approach to achieve this is to combine the output from several lasers. Such laser-clusters have been
developed and are used extensively at Lund University. The laser cluster consists of four ash-lamppumped Nd:YAG lasers. Each laser consists of a Q-switched oscillator and a single amplier. When
the Q switch is switched twice during the ash-lamp discharge, each of the Nd:YAG lasers produces
two pulses with a short time separation. The beams from the four lasers are spatially combined using
a scheme based on frequency doubling and dichroic mirrors [35]. The laser beam from the rst laser
is converted from 1064 nm into 532 nm, with a second-harmonic-generation crystal. This green beam
is then combined with the 1064-nm beam from the second laser using a dichroic mirror, reecting
532 nm and transmitting 1064 nm. The spatially overlapped beams then pass through the secondharmonic-generation crystal of laser number two. The 532-nm beam passes the crystal unaected
while the 1064-nm beam is converted into 532 nm. The same combining procedure is then repeated
until all beams are spatially overlapped. The state-of-the-art multi-YAG laser (THALES) thus produces eight pulses with a pulse energy of 400 mJ/pulse (or four pulses with 1000 mJ/pulse) at 532
nm. After pumping a dye laser/OPO and frequency doubling, up to 7 mJ of tunable radiation around
280 nm could be achieved in each of the eight pulses. The minimum time separation between the
eight pulses is 6 s (170 kHz). In order to achieve 2D LIF images, a framing camera consisting of
eight separate image intensied CCD detectors have been used (Hadland, or Specialised Imaging).
The equipment above has been used in various applications ranging from more basic understanding
of turbulent combustion [36] to IC engine applications [37].
One example where the multi-YAG cluster/Framing camera was used to perform high-speed visualization was the rst published single-cycle-resolved fuel tracer PLIF imaging performed in an HCCI
engine [38]. When running a Homogeneous Charge Compression Ignition (HCCI) engine the fuel and
air are allowed to mix prior to combustion in order to create a homogeneous, or close to homogeneous,
charge. The charge is then compressed until it auto-ignites. The following combustion occurs through
distributed reactions throughout the entire bulk volume, resulting in a very high overall combustion
rate. In Figure 12.4, a series of PLIF images is shown. It can be clearly seen how ignition is initiated
at multiple locations simultaneously. The fuel used in these measurements was ethanol. Due to the
absence of uorescence from the fuel itself, acetone, at a concentration of 10%, was added as tracer
and served as a marker of unburned regions. The excitation wavelength from the YAG-cluster was
266 nm, and uorescence was emitted in the spectral region between 350 and 550 nm with a peak
around 430 nm. The laser energy was approximately 2530 mJ in each of the eight pulses. In this
HCCI engine, the charge is completely evaporated and well mixed before ignition occurs. This feature
makes the matching of the vaporization characteristics for the fuel and the tracer less critical. Additional important characteristics are that the tracer has suitable uorescent properties and that it does
not inuence the combustion process itself [39].
Nd:YAG Laser Cluster - 3D-Visualization
Since essentially all practical combustion takes place under various degrees of turbulence, which is
truly three dimensional, there is also a need for 3D visualization. The equipment above has also been
demonstrated and applied for 3D visualization of species distributions by rapidly sweeping the laser
was approximately 25 30 mJ in each of the eight pulses. In this HCCI engine, the charge
is completely evaporated and well mixed before ignition occurs. This feature makes the
matching of the vaporization characteristics for the fuel and the tracer less critical.
Additional important characteristics are that the tracer has suitable fluorescent properties
and that it does not influence the combustion process itself [39].
255
Figure 4. A sequence of eight single-shot PLIF images showing the distributed reactions in an
Figure 12.4: A sequence of eight single-shot PLIF images showing the distributed reactions in an HCCI engine
HCCI ethanol and acetone as uorescent tracer [38].
running onengine running on ethanol and acetone as fluorescent tracer [38].
beams through the volume of - 3D-Visualization: Since essentially all practical combustion
Nd:YAG Laser Cluster interest. The rst demonstration was measurements of soot volume fractions in a turbulent ame [40]. To illustrate the technique an example of 3D fuel tracer visualization in
takes place under various degrees of turbulence, which is truly three dimensional, there is
an HCCI engine is shown in Figure 12.5. The eight laser beams from the Nd:YAG laser cluster, red
with 10 s temporal separation, are spatially separated using an oscillating mirror. This results in a
also a need for 3D visualization. The equipment above has also been demonstrated and
spatial separation of 0.5 mm between the individual laser sheets within the combustion chamber. All
applied for 3D in 70 s, corresponding to 0.5 crank angle degrees (CAD) at 1200 rpm. During this
data was collected visualization of species distributions by rapidly sweeping the laser beams
short acquisition time, the fuel conversion appears nearly frozen. To verify this assumption, images
were recorded with the same time separation but with all eight sheets overlapped at the same spatial
position in the engine. The cross section of the laser sheet was 50 0.25 mm2 . The sheet was formed
16
by a cylindrical lens ( f = 100 mm) and a spherical lens ( f = +1000 mm). To achieve parallel
laser sheets in the engine, the oscillating mirror was placed at the focal point of the spherical lens.
The uorescence was detected from beneath, through the piston extension. The camera was equipped
with proper lters to transmit the LIF signal with a maximum at 430 nm while discriminating against
scattered laser light at 266 nm and rejecting uorescence at longer wavelengths from oil residues on
the cylinder surfaces. The depth of eld of the imaging optics exceeded the dimensions of the scanned
volume. Examples of the eight individual PLIF images are shown in Figure 12.5.
A number of image processing steps have been applied to the acquired PLIF images in order to
compensate for background, absorption along the beam path, heterogeneities in the laser-intensity proles and pulse-to-pulse energy uctuations. To visualize the sequences as three-dimensional surfaces
a shape-based interpolation scheme was used. The initial step in this procedure is to produce isoconcentration curves from the eight two-dimensional images by choosing a suitable threshold. Shapebased interpolation is then used to create new iso-concentration curves, in this case seven, between the
original (measured) image planes. From these iso-concentration curves, a three-dimensional surface
is constructed and examples are shown in Figure 12.5d and 12.5e. For a more detailed description see
Ref. [41].
Laser visualization using kHz lasers
During the last few years a complement to the Nd:YAG laser cluster approach for high speed visualization has been developed. In this case a diode pumped Nd:YAG (or YLF or YVO4 ) laser allows
repetition rates up to tens of kHz. The specications of a state-of-the-art laser of this kind (Edge Wave
HD 40IV-E) show maximum pulse energies at 532, 355, and 266 nm of, 12, 5 and 3 mJ, respectively,
at a repetition rate of 10 kHz. There is also a possibility to pump a specially designed dye laser providing, at the frequency doubled wavelength required to excite the OH molecules at around 283 nm,
256
Chapter 12 M. Ald n
e
Figure 5. a) A three-dimensional fuel tracer LIF sequence. Eight parallel and equidistant 2D
Figure 12.5: A three-dimensional fuel tracer LIF sequence. Eight parallel and equidistant 2D cuts of the fuel
distribution thethe HCCI engine are shown. The data was are shown.2The data was recorded image shows
cuts of in fuel distribution in the HCCI engine recorded at CAD after TDC. b) This at 2CAD
the ves dierent grey scales corresponding to LIF intensities of <20%, 2040%, 4060%, 6080% and >80%
after TDC. b) This image shows the fives different grey scales corresponding to LIF intensities
of the maximum intensities in frame 5. c) Field of view for detection through the piston. d) Three-dimensional
of <20%, 20-40%, 40-60%, 60-80% to 50% of the maximum LIF intensity, calculated from 5. data
fuel iso-concentration surface, corresponding and > 80% of the maximum intensities in frame the c)
presented of view for detection throughThree-dimensional fuel iso-concentration surface corresponding to
Field here, z = 0 at the rst frame. e) the piston. d) Three-dimensional fuel iso-concentration
70% of the maximum LIF intensity [41].
surface, corresponding to 50% of the maximum LIF intensity, calculated from the data
presented here, z=0 at the first frame. e) Three-dimensional fuel iso-concentration surface
around 200 J at 1.5 kHz [42]. This approach has been successfully applied for high speed visualizacorresponding phenomena by several LIF intensity. [41]
tion of combustionto 70% of the maximumgroups. In these cases high speed CMOS detectors, capable
of capturing thousands of images at corresponding frame rates, have been used for recording of the
LIF images.
Laser visualization using kHz lasers: During the last few years a complement to the
As an example, Peterson and Sick performed simultaneous application been developed. In this
Nd:YAG laser cluster approach for high speed visualization has of particle image velocimetry (PIV) and planar laser-induced uorescence (PLIF) of biacetyl, used as a fuel tracer, at a rate of 4.8
case a diode pumped Nd:YAG (or YLF or YVO4) laser allows repetition rates up to tens
kHz in a spray-guided spark-ignited direct-injected (SG-SIDI) engine [43]. The aim of their study was
to understand the in-cylinder conditions that exist inlaser of this the spark plug during ignition. An
of kHz. The specifications of a state-of-the-art the vicinity kind (Edge Wave HD 40IV-E)
Nd:YLF (Quantronix Darwin Duo) dual cavity laser producing 5.9 mJ/pulse at 527 nm was used for
show measurements. An Nd:YAG (Quantronix and I) laser of, 12, 5 and 3 mJ, at 355 nm was
the PIV maximum pulse energies at 532, 355,Hawk266 nmdelivering 1.4 mJ/pulserespectively,
usedaas excitation rate of for the fuel There is alsoAa12-bit CMOSto pump a specially designed
at repetition source 10 kHz. visualization.
possibility camera (Phantom V7.1) detected
the PIV/Mie scattering and a 14-bit CMOS camera (Phantom V7.3) equipped with a lens-coupled
dye intensier (LaVision the frequency doubled wavelength required to excite the ow
image laser providing, at HS-IRO) captured the weaker biacetyl uorescence. Simultaneous OH
eld and fuel at around 283 nm, around 200 J be feasible with single crank-angle resolution in
molecules concentration measurements proved to at 1.5 kHz [42]. This approach has been
the optical SG-SIDI engine, operating at 800 rpm. A short series of four consecutive images is shown
successfully applied for high speed be seen how of combustion phenomena by several
in Figure 12.6 [43]. In this sequence it canvisualizationthe fuel spray plume impacts the spark plug.
Part of theIn these cases highthe sparkCMOS detectors, capable of the fuel impacts the ground
groups. fuel passes through speed gap, while another portion of capturing thousands of
strap and is reected downward. Velocity gradient information was also extracted from the ow eld
images at corresponding frame rates, have been used vorticity in this critical area. In short,
measurements to gain knowledge about shear strain rate and for recording of the LIF images. the
measurements show a substantial increase in equivalence ratio, velocity magnitude, shear strain rate,
and vorticity values Peterson and Sick performed simultaneous application of particle image
As an example, with large cycle-to-cycle variations during the spray event. Such results could not
have been achieved without high-speed diagnostics.
velocimetry (PIV) and planar laser-induced fluorescence (PLIF) of biacetyl, used as a

18
257
Figure 6. Simultaneous fuel concentration and velocity measurements illustrate the strong
Figure 12.6: Simultaneous fuel concentration and velocity measurements illustrate the strong coupling of spray
coupling of air-fuel mixing inside airfuel mixing inside a spray-guided Peterson and (Courtesy
momentum andspray momentum anda spray-guided DISI engine. (Courtesy of DISI engine. Sick [43].) of
Peterson and Sick [43].)
Another example of combined high-speed measurements using kHz lasers is presented by Heeger
et Anotherwho performed two-dimensional high-speed OH-PLIF and kHz lasers is presentedturbual. [44] example of combined high-speed measurements using PIV measurements in a by
lent non-premixed edge ame in an opposed-jet burner. The aim of the study was to investigate spark
Heeger et al. [44] who performed two-dimensional high-speed OH-PLIF and PIV
ignition in a turbulent non-premixed environment. Such processes are relevant e.g. for relighting
of measurements in a turbulent non-premixed edge flame in an was used for burner. The aim
gas turbine combustors. Mie scattering from aerosol droplets, opposed-jet the high-speed PIV
measurements. Excitation was provided from two diode-pumped frequency-doubled environment.
of the study was to investigate spark ignition in a turbulent non-premixed Nd:YVO4 lasers
(EdgeWave, IS4II-DE) that could generate double-pulses. The simultaneous PIV/OH-PLIF experiSuch processes are relevant e.g. rate of 2.5 kHz. gas turbine combustors. Mie scattering
ments were performed at a repetition for relighting ofFor the OH-PLIF measurements, a frequency
doubled aerosol droplets, was used for the high-speed PIV 8II-E) was operated at 2.5 kHz (3.4
from (523 nm) diode-pumped Nd:YLF laser (Edge-Wave, IS measurements. Excitation was
mJ/pulse) to pump a tuneable dye laser. The dye laser was a 10 Hz system (Radiant Dyes, NarrowSprovided from two diode-pumped frequency-doubled Nd:YVO4 lasers at 2.5 kHz was
can) modied for high-speed pumping. The maximum output pulse energy achieved(EdgeWave,
48IS4II-DE)283 nm. couldin this study CMOS cameras (LaVision, HSS5) were used for signal deJ around that Also generate double-pulses. The simultaneous PIV/OH-PLIF
tection. As for the biacetyl PLIF mentioned earlier, the PLIF camera had a lens-coupled two-stage
experiments (LaVision HS-IRO). a repetition a spherical kHz. For the diameter
image intensier were performed at After the spark, rate of 2.5ame kernel with aOH-PLIF of
about 3 mm was created. The kernel then formed a ame edge that spread radially, probably along the
20
event. These images show an apparent difference between the thickness indicated
OH-PLIF imaging and the corresponding thickness of the evaporated droplet re
the PIV images. This difference is due to the droplet evaporation in the preheat zon
258
Chapter 12 M. Ald n
e
Figure 7. Simultaneous O
(left)
and
PIV
(right)
captured in an opposed-jet
Figure 12.7: Simultaneous OH-PLIF (left)

and PIV (right) images captured in an
The images are recorded
opposed-jet burner. The images are recorded
with a frame rate offrame rate of 2.5 kHz and il
2.5 kHz and illustrate
one single ignition event. The spark is positioned at r = 4, zone single ignition event. Th
= 0. Imaged area is
25 8 mm. (Courtesy of Heeger, B hm,
o
Ahmed, Gordon, Boxx, positioned at r = 4, z = 0.
is Meier, Dreizler and
Mastorakos [44]).
area is 25 x 8 mm. (Cour
Heeger, Bhm, Ahmed, G

Boxx,
Meier,
Dreizler
Mastorakos [44]).
stoichiometric mixture fraction iso-line, as a sheet about 1.5 mm thick until it lled the be found
More state-of-the-art work using the kHz-laser approach can whole stagna- in [45, 46,
tion plane between the two nozzles of the counter ow burner. Figure 12.7 shows simultaneous OH
PLIF/PIV the of a successful ignition event. These images show visualization, between
Clearly, imagesoptimum approach for high speed an apparent dierence either a
the thickness indicated by the OH-PLIF imaging and the corresponding thickness of the evaporated
droplet as in
PIV Multi-YAG cluster the droplet evaporation in kHz zone.
pulses region inthethe images. This dierence is due to approach, or thethe preheatlaser,
few high
depends
measurement situation, the species to be measured, the size of the area to be m

More state-of-the-art work using the kHz-laser approach can be found in [45, 46, 47, 48]. Clearly,
the optimum approach for high speed visualization, either a few high power pulses as in the MultiYAG cluster approach, or the kHz laser, depends on the measurement situation, the species to be
21
measured, the size of the area to be measured and the speed required. For fuel visualization, it seems
that the available pulse energy of the kHz system may be sucient; the same may be true for OH,
probably with the exception for high pressure applications. For some other species e.g. NO and CH,
the Multi-YAG approach may be advantageous. Furthermore, if a large area and a very high speed
situation is investigated, the Multi-YAG will have an advantage. On the other hand if phenomena like
extinction or reignition in stationary ames are to be captured, the limited number of pulses and lack
of proper triggering of the Multi-YAG will favor the kHz approach.
Interestingly, there are also high speed lasers which may combine the two concepts, providing
both assets by forming a larger burst , up to hundreds of high power pulses [49, 50], which may be
very attractive, although so far these systems are not yet commercially available.
12.3
259
Thermographic phosphors
Temperature is one of the most frequently required parameters for measurements in a wide variety
of combustion applications. Besides measurements of gas temperatures, there is also a high demand
for accurate measurements of surface temperatures since this information may lead to extended and
optimized material durability and understanding of heat transfer phenomena which are of crucial
importance for various combustion apparatus, e.g. IC engines, gas turbines and furnaces.
Traditional techniques for measuring surface temperatures include those relying on thermocouples or pyrometry. However, thermocouples in combustion applications are not only intrusive but
also prone to a multitude of possible errors, including those of radiation, conduction along the wires,
catalytic eects and the slow response. Pyrometry is indeed a non-intrusive technique based on measuring the naturally emitted electromagnetic radiation by all objects. However, in combustion applications emissivity variations, spectral interferences stemming from walls and ames and absorption by
gases and soot are inevitable, making the results questionable. The present chapter describes research
directed towards an alternative measurement technique based on the use of temperature sensitive phosphors known as Thermographic Phosphors, TP, for temperature measurements in various combustion
environments as well as in related areas, e.g. sprays.
Thermographic phosphors are composed of a carrier material doped with some activator material.
The activator is often a rare earth metal. The doping concentration is typically around one percent,
which is low enough for the activator atoms to be isolated from one other by the host matrix. The host
material is microcrystalline and mostly transparent to visible radiation, i.e. it is usually the activators
that absorb and emit radiation. TPs have traditionally been used in everyday applications such as
uorescent-tube lamps, CRT-television sets and uorescent paint. Although the temperature dependence of phosphors was discovered in the 1950s, it was not until the early 1980s when Alison and
Gillies discovered the advantages of using thermographic phosphors for remote surface temperature
measurements, see [51] for an overview of TP applications up to the mid-nineties.
The luminescence properties of thermographic phosphors are very sensitive to the variations in
temperature. Most thermographic phosphors exhibit a broadband absorption in the UV region and
few more narrow absorption lines in the visible region. Several methods can be used for excitation,
such as laser radiation (pulsed or continuous), particle beams, chemical reactions or thermal excitation
in ames. For combustion applications, the excitation is usually performed by an ultraviolet or a
visible laser beam. Following excitation, the subsequent emission due to the relaxation of electrons
from the excited state to the ground state can be used for thermometry. Two methods for inferring
the temperature can be employed, the temporal and the spectral method. Both methods allow singlepoint or two-dimensional measurements to be performed. In the rst case, illustrated in Fig 12.8a, the
intensity decays exponentially and is in the simplest case single exponential in accordance with
I = I0 exp(t/),
(12.1)
where I0 is the initial emission intensity, t is time and is the lifetime time constant of the phosphorescence involved, i.e. the time taken for the intensity to decrease to 1/e of the initial emission I0 .
For some thermographic phosphors the intensity decay is not single exponential depending on
material properties. In those cases an extended model of the intensity decay can be employed to assess
all lifetimes present. The phosphorescence lifetime of the phosphor material decrease with increasing
temperature, see Figure 8a. The temperature can thus be calibrated to the lifetime or lifetimes and
retrieved by calculating the phosphorescence lifetime(s) from the measured intensity decay. This is
where I0 is the initial emission intensity, t is time and is the lifetime time constant of the
phosphorescence involved, i.e. the time taken for the intensity to decrease to 1/e of the
initial emission I0
260
Chapter 12 M. Ald n
e
(a)
(b)
Figure 8. TP- thermometry using a) the temporal decay approach (YAG:Dy), b) the
Figure 12.8: TP- thermometry using a) the temporal decay approach (YAG:Dy), b) the spectral method
spectral method (Mg3FGeO4:Mn).
(Mg3 FGeO4 :Mn).
For some by tting the intensity decay a intensity decay is not single a non-linear
normally done thermographic phosphors tothetheoretical model (e.g. Eq. 12.1), usingexponential
tting procedure. material properties. In those cases an extendedideally less the intensity decay
depending on The error in temperature in such a measurement is model of than 1% [52].
When performing the calibration procedure, a single-exponential t is often appropriate for phoscan be employed to assess all lifetimes present. The phosphorescence lifetime of the
phors that show multi-exponential features as well, i.e. the t itself does not need to be perfect as long
asphosphor material decrease with increasing temperature, see Figure 8a. The temperature
it provides robust results.
In the spectral or intensity ratio measurement method illustrated in Figure 12.8b, the fact that the
can thus be calibrated to the lifetime or lifetimes and retrieved by calculating the
various spectral emission regions for certain thermographic phosphors dier in how they are aected
by temperature, is utilized. The intensity ratio of these intensity decay. Thisindicates a temperature
phosphorescence lifetime(s) from the measured two spectral positions is normally done by
sensitivity that can be utilized for thermometry. The spectral method can be used for phosphors
fitting the intensity decay to a theoretical model (e.g. Eq. 1), using a non-linear fitting
having two emission lines that are inuenced dierently by temperature variations, some phosphors
have emissionThe error in temperature in suchby measurement is others exhibit large line shifts.
procedure. lines that are strongly broadened a temperature and ideally less than 1% [52].
Some TPs and their temporal and spectral properties for thermometry are summarized in Figure 12.9
[53]. As can be seen in most cases, the temporal approach normally gives a higher precision since the
When these curves may be steeper procedure, a single-exponential fit is temporal approach
slopes of performing the calibrationthan the spectral ratio curves. Whereas the often appropriate
has mostly been used for single point measurements, the spectral method can itself does not need
for phosphors that show multi-exponential features as well, i.e. the fit easily be extended to
two-dimensional measurements [54-59]. For this purpose, two images corresponding to the emission
to be perfect two dierent excited energy states need
originating fromas long as it provides robust results. to be acquired simultaneously. One way of
doing this is to use two ICCD detectors for registering emissions through use of a beam splitter and two
band-pass lters. Here, a calibration measurement ofmethod illustrated inICCDs is8b, the fact that
In the spectral or intensity ratio measurement the sensitivity of both Figure needed to enable
the pixel intensities in the dierent cameras to be balanced. Another method is to use an imaging
the various spectral emission regions for certain thermographic phosphors differ in with
stereoscope which uses mirrors to split the image into two identical images that can be recordedhow
a single are affected by years the development of temperature measurements using TPtwo spectral
they camera. In recent temperature, is utilized. The intensity ratio of these in connection
to combustion/high temperature applications has increased dramatically. One area of great interest is
topositions indicatesscenarios. These phenomena have for long utilized for thermometry. The
study ame spread a temperature sensitivity that can be time been studied with the aim of
improving theoretical models for ame growth and for the prediction of ame spread. Nevertheless,
the complexity of the phenomena involved makes the development of a unied model or a set of such
models a very slow process. One way of improving current models is to develop an experimental
24
database, especially of surface temperature, which plays a major role in heat exchange between the
ame and the solid material. For this purpose the fourth harmonic radiation of an Nd:YAG laser at
266 nm was used to excite the phosphor particles (Mg4 FGeO6 :Mn) deposited on/in the material of a
261
Figure 9. Temperature sensitivity for some of the phosphors surveyed in the paper.
Figure 12.9: Temperature sensitivity for some of the phosphors surveyed in the paper. Ratio calibrations are
Ratio calibrations are dashed and can be read on the y-axis at the right [53]
dashed and can be read on the y-axis at the right [53].
Related to fire studies is the pyrolysis of solid organic materials. Many parameters that
low densityplay a board (LDF) or Polymethylmethacrylate transfer models need to be validated and
ber key role in pyrolytic processes in the heat (PMMA). The temperature changes were
measured when the material was exposed to ame spread by placing it in a pool re scenario using
correctly assessed. Surface temperature is one of these parameters. In the work presented
alcohol or heptane as fuel. Both single point and two dimensional experiments were successfully
reported [60]. Ref. [61], the surface temperature of decomposing construction materials, together
in
Relatedwith their mass-loss rates during pyrolysisorganic materials. Many parameters that play a
to re studies is the pyrolysis of solid were studied.
key role in pyrolytic processes in the heat transfer models need to be validated and correctly assessed.
Surface temperature is one of these parameters. In the work presented in Ref. [61], the surface temThermographic phosphorescence has been used for measurements of surface
perature of decomposing construction materials, together with their mass-loss rates during pyrolysis
temperatures in industrial apparatus, e.g. gas turbines and IC engines. In the work
were studied.
reported in [62], laser-induced phosphorescence thermometry has been applied to in
Thermographic phosphorescence has been used for measurements of surface temperatures the industrial apparatus, e.g. gas turbines and IC engines. In the work reported in [62], laser-induced phosphorescence thermometry has been applied to the inner wall of the afterburner on a full size ghter jet
26
engine. The measurements were aiming at testing the technique under the extreme conditions prevailing in the test rig of an aviation gas turbine running at full afterburner load. The experimental set-up
is shown in Figure 12.10a.
inner wall of the afterburner on a full size fighter jet engine. The measurements were
aiming at testing the technique under the extreme conditions prevailing in the test rig of
an aviation gas turbine running at full afterburner load. The experimental set-up is shown
262
Chapter 12 M. Ald n
e
in Figure 10a.
(a)
(b)
Figure 10. a) Experimental set-up for 2D surface thermometry in a afterburner to a full
Figure 12.10: a) Experimental set-up for 2D surface thermometry in a afterburner to a full size jet engine. b)
size jet engine. b) 2D temperature map
2D temperature map.
The investigations were made on the outlet cone of the afterburner using the intensity
ratio method, and made on the outlet cone of is afterburner using the intensity ratio
The investigations were the phosphor YAG:Dy, whichthesuitable for temperature measurement method,
and the phosphor YAG:Dy, which to suitable was temperature measurement from at its third
from room temperature is 1700 K, for used. A Nd:YAG laser running room temperature to
1700 K, was used. A Nd:YAG laser running at its third harmonic, 355 nm, was used for excitation
harmonic, 355 nm, was used for excitation of the phosphor particles. A stereoscope
of the phosphor particles. A stereoscope containing interference lters for the wavelengths 455 and
containing interference filters for the wavelengths 455 and 493 nm was used for
493 nm was used for collection of signals emitted within the two spectral regions onto one half each
collection of CCD-chip. within the two spectral the resulting images were converted to
of an image intensiedsignals emittedThe intensity ratios ofregions onto one half each of an
image intensified CCD-chip. The intensity ratios of the resulting images were converted
temperatures, using predetermined calibration curves. Single-shot images of surface temperatures are
shown for an afterburner ofusing predetermined calibration curves. Single-shot images idling (no fueling of
to temperatures, both 62% and 100% load. The surface temperature at of surface
the afterburner) was 517 K, whereas, as indicated in Figure 12.10b, when the afterburner was activated
temperatures are shown for an afterburner of both 62% and 100% load. The surface
the temperature increased and leveled o at temperatures around 940 K. After the engine was operated
at full load temperature at idling (no fuelling of the afterburner) was 517 K, deactivatedindicated in
with the afterburner for 50 seconds, the afterburner was whereas, as and the temperature
Figure 10b, when the afterburner was show that at full afterburner load the standard
decreased. Statistics from the measurements activated the temperature increased and levelled off deviation,
or precision, was about 2030 K at about After theor 23%. operated at full load with the
at temperatures around 940 K. 1000 K engine was
The useafterburner for 50 seconds, the afterburner was deactivated and the both single-point and
of thermographic phosphors in internal combustion engines allows temperature
two-dimensional surface temperature measurements to be obtained. Thermocouples are a standard
decreased. Statistics from the measurements show that at full afterburner load the
tool for temperature measurement. However, due to complex geometries and the harsh environment
standard deviation, or precision, was about 20-30 K at about 1000 K or 2-3%.
in engines thermocouples are time-consuming to install. In addition, once installed, thermocouples
are dicult to shift to a dierent measurement area, which increases the complexity of an experiment
27
and the time required for it. For moving targets, e.g. pistons and valves, the instrumentation of thermocouples is even more cumbersome. Thermographic phosphors also have the advantages of enabling
two-dimensional measurements and of having high degree of applicability to diverse geometries. The
technique is also insensitive to engine vibrations or to typical pressures of around 30100 bar inside
the combustion chamber. Studies performed using either Cr3+ [63, 64] or Eu [65, 66] have not revealed
any noticeable changes in lifetime or in spectral line positions at pressures under 1GPa. Armeld et
al. [67] used TP for thermometry of the intake valve (on the stem side of the valve), and on the piston
of an SI engine [68] both two-dimensional and one-point temperature measurements in a direct injection gasoline engine using TP are demonstrated and discussed. The measurements were performed on
the intake and exhaust valves, and inside the combustion chamber on the pistons. Recently the use of
thermographic phosphors have been further developed not only for surface measurements but also for
thermometry of droplets and in sprays [69, 70] and even for measurements when seeding particles to
a gas ow [71]. In the latter case TP was used for simultaneous PIV and phosphorescence, yielding
ough a Bunsen burner flame. The 5 ns long 355 nm excitation laser pulse was formed
a sheet which penetrated the centre of the droplet. Figure 11 shows a temperature
age of a burning droplet inside the Bunsen burner flame. The short exposure time and
ong signal reduce the chemiluminescence diagnostics to negligible intensities.

interference
Recent developments in laser
263
Figure 12.11: a) A burning methanol

droplet as it was falling, photographed
with an ICCD camera. b) A temperature image of a burning methanol
droplet that was falling, recorded in the
middle of the Bunsen burner with a
camera exposure time of 10 ns [72].
Figure 11 a) A burning methanol droplet as it was falling, photographed with an

both velocities and temperatures. a burning methanol droplet that was
ICCD camera. b) A temperature image of There has also been an eort to nd phosphors with short phosphorescences lifetimes and with emission in the blue spectral region. This would facilitate measurement
falling, recorded in the middle of the Bunsen burner with a camera exposure time ofmoving objects are to be
when there is a strong continuous emission background and also when fast
captured. A further desire has been to nd phosphors with high spectral sensitivity for changes in
temperatures, which would allow precise single shot 2D measurements. As was shown in Ref [72],
ZnO:Zn and ZnO:Ga, have emission lifetimes on the nanosecond scale, emission in the blue spectral
date, basically range and very strong TP thermometry to temperatures. This has been demonstrated by recording
all applications of spectral sensitivity have been based on commercially
the ratio of two temperature emission lines from burning droplets. A series of 2D temperature meaailable phosphors with adequate performed on burning droplets of methanol [72]. A container with methanol and
spectral and/or temporal emission properties. An
surements have been
1% ZnO:Ga was heated approach would of methanol, inducing turbulence in
ernative and perhaps more appealing to the boiling pointbe to use tailor made phosphors the container and
ensuring sucient mixing of the insoluble inorganic phosphor and the methanol. The experimental
th characteristics optimized for to the one described above. The droplets were ignited by falling through a Bunsen
set up was similar each specific application. Such activities have started,
burner ame.
g. at University of Duisburg The 5 nsthe host material, Al2O3, pulse was CVD anda the sol- penetrated the
where long 355 nm excitation laser through formed to sheet which
centre of the droplet. Figure 12.11 shows a temperature image of a burning droplet inside the Bunsen
burner ame. temperature range by multiple doping with chemiluminescence interference
l method have extended its The short exposure time and strong signal reduce theCr3+ and the
to negligible intensities.
e-earth metals Dy and date, basically all applications of TP thermometry have been based on commercially available
Tm. [73]
To
phosphors with adequate spectral and/or temporal emission properties. An alternative and perhaps
more appealing approach would be to use tailor made phosphors with characteristics optimized for
each specic application. Such activities have started, e.g. at University of Duisburg where the host
material, Infrared CVD and the sol-gel method have extended its
oherent techniques -Al2 O3 , throughpolarization spectroscopy, IRPS temperature range by multiple
doping with Cr3+ and the rare-earth metals Dy and Tm [73].
10 ns [72]
References
29
[1] M. Lapp and C. M. Penney (Eds.), Laser Raman Gas Diagnostics, Plenum Press, 1974.
[2] P. R. Regnier, F. Moya and J. P. E. Taran, AIAA. J. 12, (1974) 826.
[3] M. Ald n, H. Edner, P. Grafstr m and S. Svanberg, Opt. Com. 42, (1982) 244.
e
o
[4] P. Ewart and S. V. OLeary, Opt. Lett. 11, (1986) 279-281.
[5] K. Nyholm, R. Maier, C. G. Amino and M. Kaivola, Appl. Opt. 32 (1993) 919-924.
264
Chapter 12 M. Ald n
e
[6] M. Ald n, H. Edner, G. Holmstedt, S. Svanberg and T. H gberg, Appl. Opt. 21 (1982) 1236.
e
o
[7] M. J. Dyer and D. R. Crosley, Opt .Lett. 7 (1982) 382-384.
[8] G. Kychako, R. D. Howe, R. K. Hanson and J. R. McDaniel, Appl. Opt. 21 (1982) 3225-3227.
[9] A. C. Eckbreth, Eigtheenth Symp. (Int.) on Combustion, (1980) 1471-1488.
[10] S.S. Penner, C. P. Wang and M. Y. Bahdari, Twentieth Symp. (Int.) on Combustion (1984) 11491176.
[11] R. K. Hanson, Twenty rst Symp. (Int.) on Combustion, (1986) 1677-1691.
[12] J. Wolfrum, Twenty seventh Symp. (Int.) on Combustion, (1998) 1-42.
[13] R. S. Barlow, Thirty rst Symp. (Int.) on Combustion (2006)
[14] K. Kohse-H inghaus, R. S. Barlow, M. Ald n and J. Wolfrum, Proc. Combust. Inst., 30 (2005)
o
e
89-123.
[15] A.C. Eckbreth, Laser Diagnostics for Combustion Temperature and Species, 2nd edition. Gordon
and Breach, UK, 1996.
[16] K. Kohse-H inghaus, J.B. Jeries (Eds.), Applied Combustion Diagnostics. Taylor and Francis,
o
New York, 2002.
[17] K. Kohse-H inghaus, Progress in Energy and Combustion Science 20 (3) (1994) 203-279.
o
[18] J.W. Daily, Progress in Energy and Combustion Science 23 (2) (1997) 133-199.
[19] Z.S. Li, M. Afzelius, J. Zetterberg, M. Alden, Review of Scientic Instruments 75 (10) (2004)
3208-3215.
[20] Z.S. Li, J. Kiefer, J. Zetterberg, M. Linvin, A. Leipertz, X.S. Bai, M. Alden, Proceedings of the
Combustion Institute 31 (2007) 727-735.
[21] J. Kiefer, Z.S. Li, J. Zetterberg, X. Bai, M. Ald n, Combustion and Flame 154 (2008) 802-818.
e
[22] J. Kiefer, Z.S. Li, T. Seeger, A. Leipertz, M. Ald n, Proceedings of the Combustion Institute 32
e
(2009) 921.
[23] M. Ald n, H. Edner, P. Grafstr m and S. Svanberg, Opt. Com. 42, (1982) 244.
e
o
[24] J.E.M. Goldsmith, N.M. Laurendeau, Optics Letters 15 (10) (1990) 576-578
[25] N. Georgiev and M. Ald n, Appl Spectr 51 (1997) 1229-1277
e
[26] B. Johansson, H. Neij, M. Ald n and G. Juhlin, SAE Technical Paper Series 952463, Diagnostics
e
in Diesel and SI Engines SP-1122, (1995) 129-139.
[27] M. Richter, Z.S. Li, M. Alden, Applied Spectroscopy 61 (1) (2007) 1-5
[28] A. P. Nefedov, V. A. Sinelshchikov, A. D. Usachev and A. V. Zobnin, Appl. Opt. 37 (1998) 7729.
[29] M. Ald n, S. Wallin and W. Wendt, Appl. Phys. B33, (1984) 205.
e
265
[30] T.B. Settersten, A. Dreizler, B.D. Patterson, P.E. Schrader, R.L. Farrow, Applied Physics BLasers and Optics 76 (4) (2003) 479-482
[31] J.H. Frank, X.L. Chen, B.D. Patterson, T.B. Settersten, Applied Optics 43 (12) (2004) 25882597.
[32] C.M. Vagelopoulos, J.H. Frank, Proceedings of the Combustion Institute 30 2005) 241-249.
[33] W.D. Kulatilaka, R.P. Lucht, S. Roy, J.R. Gord, T.B. Settersten, Applied Optics 46 (19) (2007)
3921-3927
[34] R.W. Schefer, W.D. Kulatilaka, B.D. Patterson, T.B. Settersten, Combustion and Flame 156 (6)
(2009) 1234-1241.
[35] C. F. Kaminski, J. Hult and M. Ald n, Appl Phys B 68 (1999) 757-760.
e
[36] C. F. Kaminski, J. Hult, M. Ald n, S. Lindenmaier, A. Dreizler, U. Maas and M. Baum, Proc
e
Comb Inst 28, (2000) 399-405.
[37] J. Hult, M. Richter, J. Nygren, M. Ald n, A. Hultqvist, M. Christensen and B. Johansson, Appl
e
Opt 41 (2002) 5002-5014.
[38] A. Hultqvist, M. Christensen, B. Johansson, M. Richter, J. Nygren, J. Hult and M. Ald n, SAE
e
Technical Paper 2002-02-0424, (2002).
[39] C. Schulz and V. Sick, Progress in Energy and Combustion Science, 31, pp. 75-121, (2005).
[40] J. Hult, B. Axelsson, A. Omrane, R. Collin, J. Nygren, P.-E. Bengtsson, M. Ald n and C. F.
e
Kaminski, Experiments in Fluids, 33 (2002) 265-269.
[41] J. Nygren, J. Hult, M. Richter, M. Ald n, M. Christensen, A. Hultqvist and B. Johansson, Proe
ceedings of the Combustion Institute 29 (2002) 679-685.
[42] I. Boxx, C. Heeger, R. Gordonb, B. B hmb, A. Dreizler, W. Meier, Combustion and Flame 156
o
(2009) 269-271.
[43] B. Peterson and V. Sick, Appl Phys B 97,(2009) 887-895.
[44] C. Heeger, B. B hm, S.F. Ahmed, R. Gordon, I. Boxx, W. Meier, A. Dreizler, E. Mastorakos,
o
Proceedings of the Combustion Institute 32 (2009) 2957-2964.
[45] B. B hm, C. Heeger, I. Boxx, W. Meier and A. Dreizler, Proceedings of the Combustion Institute
o
32 (2009) 1647-1654.
[46] R.L. Gordon, C. Heeger and A. Dreizler, Appl Phys B 96 (2009) 745-748.
[47] C. Fajardo and V. Sick, Exp Fluids 46 (2009) 43-53.
[48] J. D. Smith and V. Sick, Proceedings of the Combustion Institute 31 (2007) 747-755.
[49] N. Jiang, M. C. Webster, and W. R. Lempert, Applied Optics 48 (2009) B23-B31.
[50] B. Thurow and K. Lynch, AIAA Journal 47 (2009) 12.
266
Chapter 12 M. Ald n
e
[51] S. W. Allison and G. T. Gillies, Rev Sci Instrum 68 (1997) 2615-2650.

[52] J. P. Feist, A. L. Heyes, K. L. Choy, and B. Su, Phosphor thermometry for high temperature gas
turbine applications, IEEE, 6.1-6.7 (1999).
[53] G. S rner, M. Richrter, M. Alden, Meas. Science and Techn. 19 (2008) 125304.
a
[54] S. Alaruri, T. Bonsett, A. Brewington, E. McPheeters, and M. Wilson. Optics and letters in
engineering 31:(1999) 345-351.
[55] A.C. Edge, G. Laufer, and R.H. Krauss. Appl. Opt. 39 (2000) 546-553..
[56] D.J. Bizzak and M.K. Chyu, Rev. Sci. Instrum., 65 (1994) 101-107.
[57] R. Marino, B.Westring, G. Laufer, R. H. Krauss, and R. Whitehurst, AIAA Journal, 37(9),
September 1999.
[58] R.H. Krauss, R.G. Hellier, and J.C. McDaniel, Eu. Appl. Opt. 33 (1994) 3901-3904.
[59] G. Laufer, N.B. Rotchford, and R.H. Krauss, Am. J. Phys., 65(5), 1997.
[60] A.Omrane, F.Ossler, M.Ald n, Proceedings of the Combustion Institute 29 (2002) 2653-2659.
e
[61] A. Omrane, F. Ossler, M. Ald n, J. Svensson and J. B. C. Pettersson, Fire Mater, 29 (2005)
e
39-51.
[62] H. Seyfried, S rner. G, A. Omrane, M. Richter, and M. Ald n, ASME, GT2005-69058 (2005).
a
e
[63] Y.R. Shen and K.L. Bray, Phys. Rev. B 56 (1997) 882-891.
[64] U. H mmerich and K.L. Bray, Phys. Rev. B 51 (1995) 133-141.
o
[65] G. Webster and H.G. Drickamer, Eu. J. Chem. Phys., 72(6): (1980) 3740-3747.
[66] C.K. Jayasankar, K.S. Ramanjaneya, P. Babu, T. Tr ster, and W.B. Holzapfel, Phys. Rev. B 69
o
(2004) 2141081-2141087.
[67] J.S. Armeld, R.L. Graves, D.L. Beshears, M.R. Cates, T.V. Smith, and S.W. Allison, SAE,
971642, 1997
[68] A. Omrane, G. Juhlin, M. Ald n, G. Josefsson, J. Engstr m, T. Benham, SAE Technical Papers,
e
o
2004-01-0609 (2004)
[69] A. Omrane, G. S rner and M. Ald n, Appl Phys. B 79 ()2004 431-434.
a
e
[70] J. Brubach, A. Patt, and A. Dreizler, Appl. Phys B 83 (2006) 499-502.
[71] A. Omrane, P. Petersson, M. Ald n and M. A. Linne, Appl. Phys. B. Appl. Phys B92 (2008)
e
99-102.
[72] G S rner, M Richter, M Ald n, Opt Lett. 33 (2008) 1327-1329.
a
e
[73] C. Eckert, C. Pitsch and B. Atakan, Prog. in Organic Coatings, 67 (2010) 116-119.
Chapter 13
LDA and PIV

M.J. Tummers en L.F.G. Geers
Section 13.1 briey summarizes the basic principles of laser Doppler anemometry (LDA). For a more
comprehensive treatment of these principles, the reader is referred to standard text books such as
those by [Somerscales, 1981], [Absil, 1995] and [Albrecht et al., 2003]. Section 13.2 discusses particle image velocimetry (PIV). Details of this measurement technique can be found in, for example,
[Rael et al., 2007] and [Westerweel, 1993].
13.1
13.1.1
Laser Doppler Anemometry

Introduction
In LDA the Doppler-shift is determined of light scattered by a small particle that moves with the ow.
This Doppler-shift provides a measure for the velocity of the particle, and, therefore, for the ow
velocity. Since the introduction of the technique by [Yeh and Cummins, 1964] the use of LDA has
become widespread in both research and industrial applications. The main advantage of the technique
over conventional measuring techniques, such as hot-wire anemometry (HWA) and pressure probes,
is that it does not require a physical probe in the ow, i.e. it is a non-intrusive technique. Therefore,
the ow is not disturbed during a measurement. Other advantages of the technique are:
The Doppler frequency is a measure for the velocity component in a direction that is determined
by the geometry of the optical arrangement;
There is a linear relationship between the Doppler frequency and the velocity, resulting in a
single calibration factor. The calibration factor depends only on the geometry of the optical
arrangement and the frequency of the light source, and is independent of the ow properties;
The LDA combines a good spatial resolution with a high temporal resolution. Especially for
multi-component measurements, the spatial resolution of LDA is superior to that of HWA;
The technique is directionally sensitive which means that it is able to measure ow reversal.
The combination of the dierent items makes the technique ideally suited for measurements in turbulent ows. Both the magnitude and the direction of the instantaneous velocity vector can be measured.
It does not only give accurate information on relatively simple statistics, such as mean velocities and
Reynolds stresses, but the non-intrusive nature also enables the measurement of more complex quantities such as spatial correlation functions.
The list of advantages is impressive, it resembles the specication of an ideal measuring instrument. However, it is not dicult to list some serious drawbacks of LDA:
267
268
Chapter 13 M.J. Tummers en L.F.G. Geers

detector
(located at r=rd)
plane wave with
frequency f0
ed
spherical
wave
light source
ei
r
x
scattering particle
x
Figure 13.1: The light-scattering conguration.
LDA samples the velocity when a particle transits the measuring volume, i.e. the region in
space where the measurements are taken. Since the particles are randomly distributed in space,
the sampling times are random as well. The random nature of the sampling process precludes
the use of many standard data-processing methods (such as the fast Fourier transform (FFT)
algorithm) for the spectral analysis of the turbulent velocity uctuations;
The processing of the randomly sampled data is further complicated by the dependence of
the sampling process on the ow velocity. This phenomenon is known as the velocity bias
([McLaughlin and Tiederman, 1973]). Erroneous statistics will result if the velocity bias is ignored during the processing of the data;
LDA measures the velocity of small particles that move with the ow. Since the quantity of
interest is the uid velocity, the relationship between the particle velocity and the uid velocity
must be known.
However, the main disadvantage lies in the complexity of the measuring technique. The complexity
not only results in relatively high cost of purchasing an LDA system, it also requires an experienced
operator who is familiar with all the peculiarities of the measuring technique.
13.1.2
Basic Principles of LDA
Doppler frequency
The Doppler eect forms the basis of LDA. Light scattered by a small moving particle undergoes a
shift in frequency. This frequency shift is called the Doppler frequency and it is related to the velocity
of the particle. Below, the relationship between the Doppler frequency and the velocity is derived.
Consider Fig. 13.1 which shows a light source that generates a plane light wave with frequency
f0 . The direction of propagation of the plane wave is given by the unit vector ei . In complex notation
the plane wave is given by
Ei (x) = Ei0 ei(2 f0 tk0 ei x+i ) ,
(13.1)
where i is the initial phase and k0 = 2/0 is the wave number. The wavelength 0 is related to the
frequency of the light source as 0 = c/ f0 where c is the speed of light. A particle passing through
the plane wave scatters light in all directions and some of the light will be received by a detector. The
LDA and PIV
269
orientation of the detector is determined by the unit vector ed , and the detector is located at r = rd .
The spherical wave emitted by the particle may be represented by ([Goldstein, 1983]):
E s (r x) =
E s0 s
rx
ei(2 fs tks |rx|+s ) ,
(13.2)
where s depends on the scattering characteristics of the particle. At the particle (r = x) it follows
from Eq (13.1) and Eq (13.2) that
2 f0 t k0 ei x + i = 2 f s t + s .
(13.3)
This yields the following expression for the phase of the spherical wave at the detector (r = rd )
= 2 f0 t + i k0 ei x k s rd x
= 2 f0 t + i k0 ei x k s R ed
(13.4)
dx
t ,
dt
where it is assumed that at time t = 0 the distance between the particle and the detector is R. Furthermore, dx/dt v is the velocity vector of the particle at t = 0. The frequency of the scattered light as
seen by the detector, fw , is proportional to the time derivative of , i.e.
2 fw =
d
= 2 f0 + v(k s ed k0 ei ) .
dt
(13.5)
If the velocity of the particle is small compared to the speed of light, it can be assumed that k0 k s in
Eq (13.5), so that the frequency of the scattered light at the detector, fw , becomes
fw = f0 +
v(ed ei )
.
0
(13.6)
The second term on the right-hand side of Eq (13.6) is known as the Doppler frequency. It contains
information on the component of the velocity in the direction of the vector ed ei . This vector is
determined by the geometry of the optical arrangement of the LDA.
Heterodyne detection
It is the task of the detector to generate an output signal from which the Doppler frequency fD can be
determined. The direct measurement of the Doppler frequency requires a very high resolution of the
detector, because the Doppler frequency is much smaller than the frequency of light; typically fD / f0
108 . In the low-velocity range, say v < 300 m/s, the Doppler frequency can be determined with an
optical mixing or heterodyne technique in conjunction with a square-law detector. The essence of
heterodyning is that when two light waves with slightly dierent frequencies, fw1 and fw2 , are mixed
on the surface of a square-law detector, the output signal oscillates with the dierence frequency
fw1 fw2 . Because of its superior signal-to-noise ratio, the most widely used optical conguration is
the dual-beam heterodyne conguration. Figure 13.2 shows the optical arrangement for the dual-beam
conguration. Eq (13.6) can be applied to both incident beams, resulting in
fw1 = f0 + v(ed ei1 )/0
fw2 = f0 + v(ed ei2 )/0
v(ei1 ei2 )
.
fD fw2 fw1 =
(13.7)
270

detector
y
v
ed
lens
/2
light source
ei1
x
/2
beam
splitter
ei 2
scattering particle
Figure 13.2: The optical arrangement for the dual-beam heterodyne LDA.
The unit vectors ei1 and ei2 indicate the direction of the incident beams, and the Doppler frequency
fD is now dened as the dierence between fw1 and fw2 . Inspection of Eq (13.7) shows that fD is a
measure for the velocity component in the direction of ei1 ei2 . The Doppler frequency is independent
of the orientation of the detector with respect to the incident beams, which enables the use of large
apertures to collect more scattered light.
When a particle passes through the overlap region of the incident beams, i.e. the so-called measuring volume, it scatters light in all directions. A square-law detector, usually a photomultiplier tube,
then receives light with two slightly dierent frequencies, fw1 and fw2 . The relationship between the
output signal y(t) and the input signal x(t) of a photomultiplier is given by ([Somerscales, 1981])
y(t) =
S
Ts
t+T s /2
tT s /2
x(t )2 dt ,
(13.8)
where T s is the time constant of the photomultiplier and the constant S represents the radiant sensitivity. According to Eq (13.5) the input signal can be written as
x(t) = a1 cos(2 fw1 t + 1 ) + a2 cos(2 fw2 t + 2 ) ,
(13.9)
where 1 2 is the phase dierence between the two light waves. The phase dierence is assumed
to be constant, which indicates the need for a coherent light source such as a laser. On combining
Eqs (13.7) through (13.9) the following expression for the output signal of the photomultiplier is
obtained (assuming T s fw1
1 and T s fw2
1)
| sin( fD T s )|
1
cos(2 fD t + 1 2 ).
y(t) = S a2 a2 + S a1 a2
1 2
2
fD T s
(13.10)
For small values of the time constant T s , say T s 109 s, Eq (13.10) reduces to the following wellknown expression for the output signal of the photomultiplier
1
y(t) = S a2 a2 + S a1 a2 cos(2 fD t + 1 2 ).
2 1 2
(13.11)
The rst term on the equations right-hand side is known as the pedestal; it is the result of the spatial
distribution of the light intensity in the overlap region of both beams. The second term, called the
Doppler burst, carries the desired information, because it oscillates with the Doppler frequency fD .
LDA and PIV
271
with
frequency
shift
| fD |
without
frequency
shift
y
v
fs
ei 2
/2
x
/2
vmin
ei 1
Figure 13.3: The eect of frequency shift on

the relationship between the particle velocity
and the frequency of the photomultiplier output signal.
Figure 13.4: The interference of two plane

light waves.
Directional ambiguity
Referring to Fig. 13.2, it is easy to see that the expression for the Doppler frequency, Eq (13.7), can
be rewritten as
v(ei1 ei2 ) 2 sin(/2)
=
|v| sin ,
(13.12)
fD =
0
0
where is the angle between the unit vectors ei1 and ei2 , i.e. is the crossing angle of the incident
laser beams. Furthermore, is the angle between the velocity vector and the x-axis. The frequency
dierence fD can be positive or negative depending on the value of . However, the output of the
photomultiplier cannot distinguish between positive and negative values of fD because cos( fD ) =
cos( fD ). As a result, the LDA in its basic form is unable to determine the sign of the velocity. The
insensitivity to the direction of the particle velocity is usually referred to as the directional ambiguity.
The common method to remove this ambiguity is frequency shifting. In that case the frequency of
one of the incident beams in Fig. 13.2 is shifted by a constant value f s . This can be achieved with an
acousto-optic Bragg cell. Due to the frequency shift the relationship between Doppler frequency and
particle velocity becomes (assuming f s
f0 )
fD = f s +
2 sin(/2)
|v| sin ,
0
(13.13)
as illustrated in Fig. 13.3. If the shift frequency f s is chosen larger than the Doppler frequency that
corresponds to the smallest anticipated velocity in the ow, vmin , each value of | fD | is uniquely related
to one velocity value, and, as a consequence, the directional ambiguity is removed. In practice, one
usually sets the shift frequency f s about two times larger than the Doppler frequency that corresponds
to vmin ([Tropea, 1986]).
Fringe model
An alternative procedure to derive the relationship between the Doppler frequency and the velocity
for the dual-beam LDA, Eq (13.12), is given by the fringe model, which is due to [Rudd, 1969]. The
fringe model is often used to visualize dierent aspects of the dual-beam conguration, such as the nature of the detector output signals in case the incident beams are improperly aligned ([Durst and Stevenson, 1979]).
It also gives an interpretation of the proportionality constant between the Doppler frequency and the
272
velocity in Eq (13.12). However, the fringe model should be considered with some reserve, because
it is incorrect in the sense that it ignores the fact that heterodyning takes place on the surface of the
photomultiplier and not at the particle, see [Durst, 1982]. However, most of the predictions of the
fringe model are in accordance with the Doppler theory, as will be shown below.
If the incident beams shown in Fig. 13.4 are properly aligned, their wavefronts are nearly plane in
the overlap region, so that the light waves can be described with Eq (13.1). The intensity of the light
in the overlap region of both beams is then given by
I = (E10 + E20 )(E10 + E20 )
2
E10
2
E20
(13.14)
+ 2E10 E20 cos(2k0 y sin(/2) + 1 2 ) ,
where is the angle between the unit vectors ei1 and ei2 , y is a coordinate in the direction of ei1 - ei2
and 1 2 is the phase dierence between the two light waves. According to Eq (3.14) the intensity
varies periodically in y, and the distance between two consecutive lines of constant intensity in the
interference pattern is given by
0
df =
.
(13.15)
2 sin(/2)
The quantity d f is known as the fringe distance, and inspection of Eq (13.12) reveals that it is the
inverse of the proportionality constant between the Doppler frequency and the velocity. A small
particle passing through the interference pattern with a velocity component in the y-direction of v(=
v sin ), scatters light with an intensity that is proportional to the local value of I. The intensity of
the scattered light then oscillates with frequency
v
2 sin(/2)
=
v sin .
df
0
(13.16)
It follows from a comparison with Eq (13.12) that this is identical to the Doppler frequency.
The fringe model can also be used to visualize the eects of applying a frequency shift to remove
the directional ambiguity. If in Fig. 13.4 the frequency of one of the beams, say beam 1, is increased
with a value f s , the intensity of the light in the overlap region becomes
2
2
I = E10 + E20 + 2E10 E20 cos(2 f s t + 2k0 y sin(/2) + 1 2 ) .
(13.17)
The fringes in the interference pattern now move with velocity v s = d f f s in the positive y-direction.
As a result, a detector sees intensity variations with a frequency
fs +
2 sin(/2)
v
= fs +
v sin = fD ,
df
0
(13.18)
which is identical to Eq (13.13), the result obtained using the Doppler theory.
Amplitude bias
[Durao and Whitelaw, 1979] have shown experimentally that there is a relationship between the amplitude of a Doppler burst and the particle velocity. There study revealed that there is a tendency for
low-speed particles to produce high-amplitude Doppler bursts, and vice versa. Through this mechanism low-velocity particles have (on the average) a larger probability of being detected and validated
by the LDA signal processor than high-velocity particles. This bias towards low velocities was termed
amplitude bias by Durao and Whitelaw. They argued that the amplitude bias was due to the fact that
LDA and PIV
273
Figure 13.5: The amplitude A versus the velocity v for several values of T s ranging between 2 ns and 32 ns.
fast moving particles (on the average) spend less time in the measuring volume than slow particles.
The fast moving particles scatter less photons and, therefore, produce Doppler bursts with smaller
amplitude.
It was already shown in Section 13.1.2 (Eq (13.10)) that, in principle at least, the amplitude A of
a Doppler burst depends on the Doppler frequency through the term
A = S a1 a2
| sin( fD T s )|
,
fd T s
(13.19)
where T s is the time constant of the photomultiplier and fD is the Doppler frequency. Substitution of
Eq (13.18) in Eq (13.19) yields the following expression for A
A = S a1 a2
| sin(( f s +
( f s +
v
df
v
df
)T s )|
)T s
(13.20)
where f s is the shift frequency, v is the velocity component and d f is the fringe distance. Figure 13.5
depicts the amplitude A as a function of v for several values of the photomultiplier time constant T s
ranging between 2 ns and 32 ns. Furthermore, it is assumed that d f = 3 m and f s = 40 MHz.
Figure 13.5 shows that the amplitude A will signicantly vary with the particle velocity only when the
value of T s is large. Clearly, photomultipliers with large time constants are not suited for LDA, because such photomultipliers give rise to the amplitude bias. Figure 13.5 also shows that the amplitude
A is practically constant for small values of T s . This means that the dependence of the amplitude A
on the particle velocity can be conveniently ignored when a so-called fast response photomultiplier
is used, which is usually the case.
Particle characteristics
The light-scattering particles form an essential element of the LDA measuring system. In each application the suitability of the particles must be determined in the same way as any other element of the
274
LDA instrumentation. In general, it is highly appreciated if the particles are cheap, easy to generate,
non-corrosive and non-toxic. However, the suitability of the particles for application in LDA mainly
depends on their dynamical and optical characteristics.
The optical characteristics should be such that the particles scatter light with sucient intensity
for the photodetector to generate high-quality Doppler signals. Investigations based on Mies scattering theory, e.g. [Durst, 1982], show that the amplitude and the visibility of the Doppler signals
are dependent on the particle diameter, the refractive index of the particle material, the wavelength
of the laser light, the angle between the incident laser beams and the aperture and orientation of the
receiving optics. Generally speaking, the amplitude and visibility of the Doppler signals increase with
increasing particle size and increasing index of refraction.
The dynamical characteristics of particles determine their ability to accurately follow the uctuations in the uid velocity even at high frequencies. The motion of a rigid, spherical particle in a
viscous ow is governed by the Basset-Boussinesq-Oseen (BBO) equation, see [Somerscales, 1981].
Solutions of the BBO equation are discussed by [Hjelmfelt and Mockros, 1966]. A simplied equation of motion is given by (see [Somerscales, 1981])
p
d3 du p
p
6
dt
= 3 f d p (u f u p ),
(13.21)
where is the uid kinematic viscosity, d p is the particle diameter, u p and u f are the particle and uid
velocities and p and f are the particle and uid densities, respectively. The equations left-hand
side represents the force to accelerate the particle. The term on the right-hand side is the drag of the
particle for which Stokes drag law is used. The validity of Eq (13.21) is restricted to large values of
the density ratio r p / f and not too large acceleration. Furthermore, the eects of, for example,
centrifugal forces, electrostatic forces and gravity are ignored.
Eq (13.21) will be used here to formulate a criterion for the diameter of the particles. Following
[Hjelmfelt and Mockros, 1966] the uid velocity and the particle velocity are expressed in terms of
Fourier components. Substitution of u f = eit and u p = ()eit in Eq (13.21) yields the amplitude
ratio || as
18
|()| =
.
(13.22)
with
=
r d2
2 + 2
p
The amplitude ratio can be interpreted as a measure for the sensitivity of the particles to changes in the
uid velocity. It is seen from Eq (13.22) that the particle motion is attenuated at high frequencies. The
maximum diameter of a particle that follows the velocity uctuations up to 1 kHz, 5 kHz and 10 kHz
for || = 0.99 can be determined from Eq (13.22). For a number of frequently used seed materials
the thus obtained diameters are listed in Table 13.1. From these results it can be concluded that oil
particles with a diameter of typically 1 m accurately track the velocity uctuations in low-speed air
ows. High-speed ows generally require smaller particles, because of the energy of the velocity
uctuations at higher frequencies. The presence of a shock wave in a supersonic ow provides a
further motivation to use submicron particles. Due to the strong decelerations across the shock wave
the particle velocity lags the uid velocity. This phenomenon has been studied by [Yanta et al., 1971]
and more recently by [Maurice, 1992]. The particle lag may result in a severe overestimation of the
mean velocity and the turbulence intensity at locations directly downstream of the shock wave if too
large particles are used. In general, particles that accurately follow the abrupt velocity changes in
supersonic ows should have diameters less than 0.3 m.
There are ows with practical relevance for which Eq (13.21) is invalid. In vortical ows the centrifugal forces induce a migration of particles away from the core region (for r > 1), thus reducing
LDA and PIV
275
seed
material
silicone oil
rizella oil
polystyreen
teon
titanium oxide
density ratio r
(in air)
620
711
865
1800
3500
maximum diameter m
1 kHz 5 kHz 10 kHz
3.1
1.4
1.0
2.9
1.3
0.9
2.6
1.2
0.8
1.8
0.8
0.6
1.3
0.6
0.4
Table 13.1: Maximum particle diameter for various seed materials.

Band Pass
Filter
Mixer
Low Pass
Filter
Sampler
FFT
Processor
Curve Fit &

Validation
Low Pass Filter
+ Rectifier
Gain
Host
Computer
envelope
Low Pass
Filter
pedestal
Burst
Dection
arrival time
transit time
Figure 13.6: The basic principles of the Dantec BSA processor.
the particle concentration in the core. The particle concentration may become so low that LDA measurements in the core become almost impossible as reported by [Meyers and Hepner, 1988]. Details
on the particle motion in vortical ows can be found in e.g. [Dring and Suo, 1978].
Signal processor
The principle task of a signal processor is to extract the Doppler frequency (i.e. velocity) from the
photomultiplier output signal. Usually, the signal processor also measures other quantities such as
the arrival time of the particles and the duration of the Doppler bursts, i.e. the transit time of the
particles. Two commonly used signal processors for sparsely seeded ows are counter processors
and spectrum analyzers. A detailed discussion of the characteristics of the two types of processors
and a comparison between their performances is beyond the scope of this thesis. Instead, this section
describes only the basic principles of one representative of the latter type of signal processor. The
processor to be described is the Burst Spectrum Analyzer (BSA) which is manufactured by Dantec,
and available since the late 1980s.
The BSA processor performs a spectral analysis of the bandpass-ltered output signal of the photomultiplier. The Doppler frequency then follows from the location of the peak in the computed power
spectrum. The basic principles of the Dantec BSA are illustrated in Fig. 13.6. The output signal of the
photomultiplier is rst amplied by a factor set by the operator and then bandpass ltered to remove
frequency components outside the anticipated range of Doppler frequencies. A burst detection scheme
determines whether the bandpass ltered signal contains a Doppler burst or not. Burst detection can
276
be based on the pedestal or on the so called envelope. The envelope is obtained by rectifying and
low-pass ltering of the bandpass ltered signal. When the envelope exceeds a 25 mV threshold,
the sampler and the transit time counter are started and the arrival time is measured. The sampler is
restarted each time the envelope exceeds the next higher threshold level (50 mV, 75 mV etc.). This is
done to ensure that the samples are taken from the central part of the Doppler burst. The transit time
counter is stopped when the envelope decreases below 12.5 mV.
While the burst detection scheme is carried out, the bandpass ltered signal is led through a
mixer unit which shifts the power spectrum by a value of fc towards lower frequencies. The aim
of this shift is to increase the resolution of the computed power spectrum. The centre frequency fc
is selected by the operator in conjunction with the bandwidth Bw so that the cut-o frequencies of
the bandpass lter are given by fc Bw /2. The down-shifted signal is low-pass ltered and then
sampled at regular time intervals, tsam . The number of samples nrec is called the record length, and
its value can be set by the operator at 8, 16, 32 or 64. The inverse of the time interval tsam is called
the sampling frequency fsam . The resolution of the computed spectrum is proportional to fsam /nrec ,
which reduces to 1.5Bw /nrec because the BSA has a xed relationship between the sampling frequency
and the bandwidth: fsam = 1.5Bw . A hardwired FFT processor then computes a spectrum from the
samples.
In the next step a sinc function is tted to the computed spectrum at the frequency with the highest
peak and its neighbouring frequencies. The Doppler frequency follows as the frequency for which the
sinc function achieves a maximum. The thus determined Doppler frequency is validated by means of a
comparison between the two highest peaks in the spectrum. The Doppler frequency is validated if the
primary peak of the spectrum exceeds the secondary peak by a factor of 4 or higher. After validation
the Doppler frequency together with the arrival time (optional) and the transit time (optional) are
transferred to a computer.
Unlike, for example, TSI counter processors, the BSA processors cannot carry out a time-coincidence
test to ensure that the measured Doppler frequencies originate from the same particle in case of a
multi-component measurement. However, the time-coincidence test can be performed in the software
that is used to reduce the raw data, provided that for each processor the arrival times of the particles
are stored on disk.
13.2
Particle Image Velocimetry
Particle Image Velocimetry, or PIV, is a whole eld velocity measurement technique. In other words,
instantaneous velocities are measured in one plane in a ow at an instant in time. Small tracer particles
added to the ow are illuminated in a plane at least twice within a short time interval. The light
scattered by the tracer particles is recorded by a camera either on a single frame or on a sequence of
frames. By determining the displacement of particle patterns during the time interval velocities can
be evaluated.
13.2.1
Principles of the technique
Tracer particles
The requirements for the seeding particles for PIV are generally equal to those for LDA. Firstly, they
need to follow the ow faithfully, which means they should have a lag time smaller than the typical
ow time scales. Secondly, they must scatter light with sucient intensity. Particles crossing the light
sheet in the ow scatter the light that is subsequently recorded by a CCD sensor. Since the obtained
particle image intensity and therefore the contrast of the PIV recordings is directly proportional to the
LDA and PIV
277
scattered light power, it is often more eective and economical to increase the images intensity by
properly choosing the scattering particles than by increasing the laser power.
Light source
The particles are illuminated by a pulsed light sheet. The duration of the pulse should be short with
respect to the typical timescales associated with the ow to avoid motion blurring of the particle
images. During this short pulse the light scattered from the small tracer particles will expose the CCD
sensor, implying a high power light source. In most PIV systems Nd:YAG lasers (frequency doubled,
so = 532 nm) are used for this task, because they can provide pulses with a duration in the order of
several nanoseconds with constant pulse energy. An additional advantage of the use of lasers is their
mono-chromaticity allowing easy collimation of the light beam.
Because Nd:YAG lasers have a typical repetition rate of 10 to 15 Hz while in PIV the time interval
between pulses is typically a few microseconds, two lasers are required. To this end the dual cavity
Nd:YAG lasers were developed. These lasers have two laser cavities of which the beams are combined
with precision optics. Figure 13.7 shows a schematic representation of the dual cavity laser.
The light from the laser is passed through a combination of lenses that transform the beam into a
sheet. One example of a possible lens combination is sketched in Figure 13.8. A concave cylindrical
lens is used to create the sheet from the laser beam. A subsequent convex spherical lens can be used
to control the sheet thickness and at the same time it limits the spreading of the sheet.
Diraction limited imaging
The particles can be considered point sources of light because of their small size. The projection of
a point source onto a screen or image sensor via a lens yields a Fraunhofer diraction pattern known
as the Airy disk. This is the impulse response function of an aberration-free lens. The diameter of
the Airy disk d s represents the smallest particle image (i.e. pulse response) that can be obtained for a
given imaging conguration [Rael et al., 1998]:
d s = 2.44 f # (M + 1) ,
Rear mirror
quarter waveplate
Pockels cell
Output coupler
Pump cavity
(13.23)
Dichroic mirrors
half waveplate Dielectric

polarizer
Lens
Doubling
crystal
Figure 13.7: Sketch of a typical dual cavity Nd:YAG laser.
278

Cylindrical lens Spherical lens
Cylindrical lens
(a) side view.
Spherical lens
(b) top view.
Figure 13.8: Sketch of a possible conguration for the sheet optics.
where f # is the numerical aperture of the imaging lens (i.e. the ratio of the focal length f and the
aperture diameter Da ), M is the image magnication factor, and is the wavelength of the light. The
diameter of the particle image on the recording sensor d can be estimated using:
d =
(Md p )2 + d2 .
s
(13.24)
For a typical PIV system using a Nd:YAG laser (= 532 nm), a numerical aperture of 2.8, and a
magnication factor of 0.1 the diameter of the Airy disc is 4.0 m. This means that the image diameter
of 1 m particles is also 4m. The image diameter of the particles is therefore controlled by the optics,
explaining the term diraction limited imaging. Because the magnication factor is determined by the
sizes of the desired eld of view and the imaging sensor, and the wavelength of light is xed by the
choice of laser, the particle image size can be controlled by the numerical aperture. By reducing
the aperture diameter the numerical aperture will increase and the particle images will be larger, but
the intensity of the light collected by the imaging sensor will decrease yielding images with lower
contrast.
Image acquisition
Most modern PIV systems use Charge Coupled Device (CCD) cameras for the acquisition of particle
images. A CCD chip consists of small semi-conductor cells (typically 7-10 m) called pixels that can
convert light into electric charge. A modern CCD chip contains typically one to four million pixels.
CCD chips need to be cooled to keep the noise generated by thermal eects in the semi-conductor
below a certain level. Read-out noise forms another source of error. This limits the number of images
that can be acquired per unit of time, because this type of noise increases with the read-out frequency
of the chip. The typical CCD sensor has a maximum image acquisition rate of about 10 to 15 images
per second.
In single exposure double frame PIV a camera is needed that can acquire two images within a time
interval of several microseconds. A special kind of CCD camera was developed for this application.
In these cameras an image buer integrated on the CCD chip temporarily stores the rst image while
the second image is acquired. After the acquisition both images are stored in a computer for the
calculation of velocities.
Data reduction
For the calculation of velocity distributions from PIV images, the images are subdivided into small
regions called interrogation areas. In these areas it is assumed the velocity of the ow is uniform, so
the particle pattern in the interrogation area will not be distorted during the time interval between the
LDA and PIV
279
two exposures. Each interrogation area in the rst image is correlated with the corresponding area
in the second image, as illustrated in Figure 13.9. The calculation of the cross-correlation is done in
a fast and computationally ecient way according to the correlation theorem [Priestley, 1981] using
Fast Fourier Transformation (FFT). First, both interrogation areas are Fourier transformed (denoted
by FT in the gure). Next, the transformed area of image 1 is multiplied by the complex conjugate of

the transformed area of image 2, i.e. I I is computed. The quantity I I is then inverse transformed
1 ). The resulting correlation function is shown in Figure 13.10. R is the displacement correlation
(FT
D
peak. This peak represents the cross-correlation between images of the same particles in the rst and
the second image. RC is the convolution of the mean image intensities and RF is the uctuating
noise component. The location of the displacement correlation peak with respect to the origin of the
correlation function represents the mean in-plane particle displacement vector s from which the local
mean velocity vector v can be determined using:
v=
s
,
t
(13.25)
where t is the time separation between the two images.

Criteria for accurate velocity measurements
The height and the width of correlation peaks depend on the particle image diameter, the number of
particle images present in the interrogation area in both images, and the velocity gradient over the
interrogation area. To be able to determine the velocity in an interrogation area suciently accurate a
number of criteria were dened with regard to the particle image pair density and the velocity gradient
over an interrogation area:
The particle image pair density is the number of particle image pairs per interrogation area, i.e.
the number of particles of which images can be found in corresponding interrogation areas in
both images. It can be calculated from the concentration of particle images s in one interrogation
area, NI , the fraction of particle images lost due to in-plane displacement, (1 F I ), and the
fraction lost due to out-of-plane displacement, (1 FO ). F I and FO are equal to one if no
pairs were lost and zero if all the pairs are lost. The particle image pair density NI F I FO should
exceed 5 to assure a probability of 95% for the detection of at least two particle image pairs
[Keane and Adrian, 1992].
To achieve minimum in-plane loss of pairs (i.e. F I 1) the dimensionless maximum displacement of the particle image pattern should not exceed 0.25:
M|vmax |t
< 0.25 ,
dI
(13.26)
Figure 13.9: Schematic representation of the cross-correlation of interrogation areas [Rael et al., 1998].
280
Figure 13.10: An example of a cross-correlation function [Rael et al., 1998].
where vmax is the maximum in-plane velocity and dI is the width of an interrogation area on the
image plane. For the out-of-plane displacement a similar criterion was dened (FO 1):
|wmax |t
< 0.25 ,
z0
(13.27)
where wmax is the maximum out-of-plane velocity and z0 is the thickness of the laser sheet.
According to [Keane and Adrian, 1990] the width and the height of the displacement correlation
peak strongly depend on the velocity gradient over the interrogation area. In case of strong
velocity gradients the peak broadens and its amplitude is diminished, decreasing the likelihood
of detecting the displacement peak. This eect is called detection bias. Acceptable detection
probability exists if
Mut
1,
(13.28)
d
where u is the velocity dierence over the interrogation area as a consequence of the velocity
gradient.
A second eect of velocity gradients is the gradient bias. It occurs because the correlation is
biased towards the faster moving particle within the measurement volume. This means that the
correlation peak is statistically biased towards velocities lower that the average velocity in the
interrogation area. The criterion that needs to be satised to minimize this eect is
Mut
< 0.05 .
dI
(13.29)
Displacement estimation
After cross-correlating interrogation areas in the two images the displacement correlation peak is
detected by searching for the maximum of the correlation function. Due to the discretization of the
function the location of the correlation peak can only be determined in integer numbers of pixels
with an uncertainty of 0.5 pixels. This can be remedied by using a subpixel displacement estimator.
This estimator uses pixels in the direct neighborhood of the correlation peak to make a more accurate
estimate of its position. Because correlation peaks can be narrow, only three pixels are taken into
account, hence the name three-point estimators.
LDA and PIV
281
y/Ds
0.5
0.25
0.25
x/Ds
0.5
Figure 13.11: Example of raw velocity data resulting from PIV measurements in a turbulent jet.
Most three-point estimators t a Gaussian function through the three pixels. The particle images
have an Airy shaped intensity distribution that can be approximated very well by a Gaussian. The
correlation between two Gaussian functions is also Gaussian, so the best t for the correlation function
is a Gaussian. The Gaussian peak estimator is dened by:
ln Ri1, j ln Ri+1, j
,
2 ln Ri1, j 4 ln Ri, j + 2 ln Ri+1, j
ln Ri, j1 ln Ri, j+1
y0 = j +
,
2 ln Ri, j1 4 ln Ri, j + 2 ln Ri, j+1
x0 = i +
(13.30)
(13.31)
where x0 and y0 are the precise coordinates of the peak, i and j are the integer coordinates of the peak,
and Ri, j is the value of the correlation function at (i, j).
If the size of the particle images is smaller than or equal to the size of one pixel, the correlation
peak will be very narrow. As a consequence, a bias error is introduced in the three-point estimator. The
resulting velocity data will be biased towards integer displacements yielding a multi-modal velocity
histogram. This eect is known as peak locking and it can be avoided by adjusting the particle image
size (e.g. by choosing another camera lens or increasing the numerical aperture). The optimum
particle size is around 2 pixels [Willert, 1996].
Data validation
After the analysis of the PIV images some of the velocity vectors are obviously incorrect, because
their magnitude and direction strongly deviates from those of their direct neighbors. These spurious
vectors result from the incorrect detection of displacement correlation peaks, because they are of the
same order of magnitude as the noise in the correlation function. Spurious vectors are often found in
regions where one or more of the above mentioned criteria is not satised or near walls, where wall
reection of the incident laser light dominates the image. Figure 13.11 shows an example of a vector
plot of raw velocity data in a turbulent jet.
There are several statistical methods to detect the spurious vectors and remove them from the data
set [?, for details see]]Westerweel:1994. Of these methods the local median lter is the most popular
282
technique. For each vector a 33 neighborhood is selected and the median is calculated of the eight
vectors surrounding the center vector. The median value of a set of vectors is dened here as either
with respect to vector length or with respect to its individual components. The dierence between the
median vector and the center vector should remain within a certain limit:
vi, j vsurr(i, j) < surr(i, j) .
(13.32)
where vi, j is the velocity vector on position (i, j) in the grid, vsurr(i, j) is the median of the vectors
surrounding the one at (i, j), surr(i, j) is the standard deviation of their magnitudes, and is a parameter
determined on the basis of a condence interval. Because surr(i, j) is not known, the sample standard
deviation ssurr(i, j) can be used. This method is similar to the median ltering technique from image
processing.
The probability of nding a spurious vector in the direct neighborhood of another spurious vector
is given by a binomial distribution. Usually the aim in a PIV experiment is to have a valid data yield
of 95% or more (i.e. less than 5% spurious vectors). In that case the probability of nding two or
more neighboring spurious vectors is less than 14.9% [Westerweel, 1994].
After detection the spurious vectors are discarded and the empty spaces lled with vectors resulting from a bilinear interpolation procedure of the neighboring vectors. An additional advantage of
bilinear interpolation is that it fullls continuity [Rael et al., 1998].
13.2.2
Advanced techniques
Adaptive cross-correlation
The velocities resulting from PIV measurements are aected by measurement noise that is due to the
nite number of particle images within the interrogation area and the number of pixels by which the
interrogation area is discretized. [Westerweel et al., 1997] have shown that the measurement error due
to this kind of noise is practically independent of the displacement, except for small displacements
(i.e. 1 px) where the measurement error is directly proportional to the displacement. This can be used
2
to improve the precision of PIV measurements.
First the integer displacements are estimated using the cross-correlation procedure described above.
After ltering out the spurious vectors and interpolating the empty spaces, the procedure is repeated,
but now the interrogation areas in the second image are oset with respect to the rst image by the
local integer displacement. Eectively more particle image pairs are found, so F I 1. Typically,
the adaptive correlation has to be done twice to obtain convergence of the oset vectors. The residual
1
displacement is now between 2 px and 1 px, ensuring low measurement error.
2
[Westerweel et al., 1997] show that the signal-to-noise ratio becomes independent from the uctuating displacement for low turbulence ows, while there is a threefold improvement of the signalto-noise ration for high turbulence ows. Additionally, the maximum measurable velocity of the PIV
measurement no longer depends on the size of the interrogation area, increasing the dynamic range.
Grid renement
In order to study ows on a range of length scales that is as wide as possible, the grid on which
the the velocity vectors are determined should be as ne as possible. However, standard evaluation
schemes cannot use small interrogation areas without losing the correlation signal due to the larger
displacements. This can be solved by integrating grid renement and adaptive cross-correlation.
LDA and PIV

Operator
283
Implementation
Accuracy
Uncertainty
Forward dierence
du
dx i+1/2
ui+1 ui
x
O(x)
u
1.41 x
Backward dierence
du
dx i1/2
ui ui1
x
O(x)
u
1.41 x
Centered dierence
du
dx i
ui+1 ui1
2x
O(x2 )
u
0.71 x
4th order Richardson
du
dx i
ui2 8ui1 +8ui+1 ui+2

12x
O(x4 )
u
0.95 x
Least squares
du
dx i
2ui+2 +ui+1 ui1 2ui2

10x
O(x2 )
u
0.32 x
Table 13.2: Five examples of dierence schemes for the estimation of velocity gradients [?, after]]RafWilKom:1998.
First a cross-correlation procedure is conducted using large interrogation areas (128128 or 6464)
that do not overlap each other. Because Fourier transforming large interrogation areas is time consuming, it is also possible to de-sample the images and take smaller interrogation areas. Next the spurious
vectors are detected and substituted by interpolated vectors. The resulting vector eld is interpolated
to a ner grid and the vectors are used as interrogation area osets for an adaptive cross-correlation
on the ner grid. After substituting the spurious vectors again the procedure can be repeated until the
desired resolution is obtained. In the last step it is also possible to use overlapping interrogation areas
to increase the resolution even further. This procedure is called the Window Displacement Iterative
Multigrid (WIDIM) method and it is discussed in detail by [Scarano and Riethmuller, 1999].
Dierential quantities
Many terms in the Navier-Stokes equation and the k-equation, and some structure identication methods, are based on spatial derivatives of the velocity eld structure identication methods. The calculation of these derivatives implies a certain continuity and smoothness of the velocity eld. Because
the velocity eld is discrete, it needs to be interpolated by tting an analytical function through the
data or by using one of the many discrete dierential operators available to estimate the derivatives.
Table 13.2 shows a few possible schemes with their accuracy and the measurement uncertainty. x is
the grid spacing, and u is the measurement noise in the u-component of the velocity
In discrete derivative computation a compromise needs to be made between the order of the lter,
the number of points used for the derivation, the frequency response and the noise amplication.
Generally, for classical nite dierence schemes, the order increases with the number of points used
for the computation. The frequency response increases when the truncation error decreases, which is
directly linked to the order of the lter [Foucaut and Stanislas, 2002]. However, noise amplication
also increases with the order of the lter, so there is an optimum for the order of the lter. By means
of spectral analysis [Foucaut and Stanislas, 2002] show that a second-order central dierence scheme
is enough to obtain a good accuracy of the derivative.
The grid in PIV data is not necessarily equidistant in space, because of perspective projection for
instance. Another possibility is that the grid is rotated with respect to physical coordinates. Figure
13.12 shows both situations. If the derivatives are calculated with respect to the grid of the image, the
grid coordinates need to be transformed in order to have accurate estimates of the derivatives with respect to physical coordinates. For velocity component u at position (i, j) in the grid the transformation
284

PIV image
PIV image
Physical coordinate system
Physical coordinate system
(a) perspective projection
(b) rotation
Figure 13.12: Alignment mismatch for PIV measurements
is as follows:
u
x
u
y
i, j
x
i
x
j
y
i
y
j
i, j
u
i
u
j
= J i, j
i, j
u
i
u
j
(13.33)
i, j
where x and y are physical coordinates, i and j are grid coordinates, and J i, j is the Jacobian on
position (i, j). The elements of the Jacobian can be calculated from the coordinate mapping that was
made during the measurements by acquiring an image of a grid with known spacing. The derivatives
with respect to the grid coordinates are estimated using a nite dierence scheme.
As was mentioned above, the resolution of PIV can be increased by using overlapping interrogation areas. Because this causes neighboring velocity vectors to be correlated, the estimates of the
derivatives will also be aected in some cases. [Rael et al., 1998] show that if the interrogation areas
are 50% overlapping, the central dierence scheme will not be aected, because data are used from
non-overlapping areas.
References
[Absil, 1995] Absil, L. (1995). Analysis of the laser doppler measurement technique for application
in turbulent ows. Phd. thesis, Delft University of Technology.
[Albrecht et al., 2003] Albrecht, H., Borys, M., Damaschke, N., and Tropea, C. (2003). Laser
Doppler and Phase Doppler Measurement Techniques. Springer.
[Dring and Suo, 1978] Dring, R. and Suo, M. (1978). Particle trajectories in swirling ows. Journal
of Energy, 2(4):233237.
[Durao and Whitelaw, 1979] Durao, D. and Whitelaw, J. (1979). Relationship between velocity and
signal quality in laser doppler anemometry. J. Phys. E:Sci. Instrum., 12:2750.
[Durst, 1982] Durst, F. (1982). Combined measurements of particle velocities, size distributions, and
concentrations. Journal of Fluids Engineering, 104:285296.
[Durst and Stevenson, 1979] Durst, F. and Stevenson, W. (1979). Inuence of gaussian beam properties on laser doppler signals. Applied Optics, 18(4):516524.
LDA and PIV
285
[Foucaut and Stanislas, 2002] Foucaut, J. M. and Stanislas, M. (2002). Some considerations on the
accuracy and frequency response of some derivative lters applied to particle image velocimetry
vector elds. Meas. Sci. Technol., 13:10581071.
[Goldstein, 1983] Goldstein, R. (1983). Fluid Mechanics Measurements, chapter 5. Hemisphere
Publishing Company.
[Hjelmfelt and Mockros, 1966] Hjelmfelt, A. and Mockros, L. (1966). Motion of discrete particles
in a turbulent uid. Appl. Sci. Res., 16:149161.
[Keane and Adrian, 1990] Keane, R. D. and Adrian, R. J. (1990). Optimization of particle image
velocimeters part I: Double pulsed systems. Meas. Sci. Technol., 1:12021215.
[Keane and Adrian, 1992] Keane, R. D. and Adrian, R. J. (1992). Theory of cross-correlation analysis
of piv images. Appl. Sci. Res., 49:191215.
[Maurice, 1992] Maurice, M. (1992). Aiaa-92 paper 0764. In AIAA.
[McLaughlin and Tiederman, 1973] McLaughlin, D. and Tiederman, W. (1973). Biasing correction
for individual realization of laser anemometer measurements in turbulent ows. Phys. Fluids, 16:7.
[Meyers and Hepner, 1988] Meyers, J. and Hepner, T. (1988). Measurement of leading edge vortices
from a delta wing using a three component laser velocimeter. In Proceedings of the AIAA 15th
Aerodynamic Testing Conference.
[Priestley, 1981] Priestley, M. B. (1981). Spectral Analysis and Time Series, volume 1. Academic
Press Inc., New York.
[Rael et al., 1998] Rael, M., Willert, C., and Kompenhans, J. (1998). Particle Image Velocimetry
A Practical Guide. Springer-Verlag, Berlin, 1 st edition.
[Rael et al., 2007] Rael, M., Willert, C., Wereley, S., and Kompenhans, J. (2007). Particle Image
Velocimetry - A Practical Guide. Springer, 2nd edition.
[Rudd, 1969] Rudd, M. (1969). A new theretical model for the laser doppler meter. J. Phys. E., 2:4.
[Scarano and Riethmuller, 1999] Scarano, F. and Riethmuller, M. L. (1999). Iterative multigrid approach in PIV image processing with discrete window oset. Exp. Fluids, 26:513523.
[Somerscales, 1981] Somerscales, E. (1981). Fluid Dynamics, volume 18 of Methods of Experimental Physics, chapter 1. Academic Press.
[Tropea, 1986] Tropea, C. (1986). A practical aid for choosing the shift frequency in lda. Exp. Fluids,
4:7980.
[Westerweel, 1993] Westerweel, J. (1993). Digital Particle Image Velocimetry - Theory and Application. PhD thesis, Delft University of Technology, Delft.
[Westerweel, 1994] Westerweel, J. (1994). Ecient detection of spurious vectors in particle image
velocimetry data. Exp. Fluids, 16:236247.
[Westerweel et al., 1997] Westerweel, J., Dabiri, D., and Gharib, M. (1997). The eect of a discrete
window oset on the accuracy of cross-correlation analysis of digital PIV recordings. Exp. Fluids,
23:2028.
286
[Willert, 1996] Willert, C. (1996). The fully digital evaluation of photographic PIV recordings. Appl.
Sci. Res., 56:79102.
[Yanta et al., 1971] Yanta, W., Gates, D., and Brown, F. (1971). The use of a laser-doppler velocimeter in supersonic ow. In Proceedings of the Second International Workshop on Laser Velocimetry,
volume AIAA paper 71-287.
[Yeh and Cummins, 1964] Yeh, Y. and Cummins, H. (1964). Localized uid ow measurements with
a he-ne laser spectrometer. Appl. Phys. Lett., 4:3.
Chapter 14
Thermometry by coherent anti-Stokes Raman

spectroscopy
E.H. van Veen
14.1
Introduction
In combustion, temperature is an important variable to be measured. Thermocouples have been traditionally used, but they have the drawback that their accuracy for high temperatures is rather low.
Radiation losses, convection and conduction corrections and catalytic properties of the thermocouple
material are error sources. Besides they are intrusive, limited in spatial resolution, and not able to
measure temperature uctuations in turbulent ames. Several laser techniques can be applied for temperature measurements. The technique of choice for precise and accurate thermometry in gas-phase
(reacting) ows is Coherent Anti-Stokes Raman Spectroscopy (CARS) [1].
CARS has been an attractive diagnostic technique for various combustion processes for over 30
years since the pioneering work of Jean-Pierre Taran et al [2]. It allows nonintrusive, temporally
and spatially resolved measurements of temperature. Sometimes, relative concentrations of major
species can be determined. The large increase in SNR aorded by CARS relative to spontaneous
Raman scattering makes it possible to probe the highly luminous, particle laden ames as used in
many practical combustion systems. In air-fed turbulent ames, broadband CARS thermometry of
the nitrogen molecule is most commonly applied to meas-ure instantaneous and mean temperatures,
temperature uctuations and PDFs (probability density functions).
14.2
Spectroscopy
Four-photon processes with a Raman resonance is now a classical scheme of nonlinear, optical spectroscopy. In particular, CARS is a third-order nonlinear, four-wave mixing process, wherein a coherent, well-dened Raman-induced eld is created (Fig. 14.2). A nascent signal at a new frequency is
generated by the interaction of the electric elds of three intense laser beams with molecules in the
probed medium. Two laser beams, one pump beam of frequency P and one Stokes beam at S are
tuned so that the dierence P S is resonant with a rovibrational Raman-active transition at ,J
in the molecule of interest. Typically, a transition in the Q-branch is probed, for which the vibrational quantum number changes by 1 ( = 1) and the rotational quantum number doesnt change
(J = 0). At each point in the medium, the phase of the driven vibration depends upon the phase of
the two input waves. This driven mode can eciently scatter a third probe wave, a laser beam also at
P , if the wave vectors of the probe and vibration add to give the wave vector of the scattered wave.
A coherently oscillating polarization is produced radiating a laser-like CARS beam at the anti-Stokes
287
288
Chapter 14 E.H. van Veen
Figure 14.1: Energy level scheme for CARS (not to scale).
frequency AS = 2P S . The strength of the interaction is related to the third-order susceptibility

of the medium, which is strongly enhanced when the frequency dierence P S matches one of
the Raman shifts of the medium. The intensity of the CARS signal at frequency AS is given by
IAS =
2
AS
(3)
I 2 IS CARS
n2 nS nAS c4 2 P
P
0
sinc
k
2
(14.1)
with IP , IS , IAS : beam intensities at frequencies P , S , AS respectively; ni : refractive index; c:

speed of light; 0 : permittivity of free space; l: length of probe volume; k: phase mismatch of
CARS beam and laser beams. The signal intensity is proportional to the Stokes laser intensity and the
square of the pump laser intensity, and, therefore, is greatly enhanced with the use of pulsed lasers.
(3)
The third-order susceptibility of the medium CARS is a summation of the resonant and nonresonant
susceptibility contributions. The nonresonant part is associated with the fast response of the electrons
of the medium to the incident elds. The resonant part is related to the slower nuclear responses, and,
retaining only terms explicitly containing ,J P S , is given by
(3)
CARS =
82 nP 0 c4 NN,J
nS 4
S
,J P + S
,J
j 2
(14.2)
,J
with N: number of molecules; N,J : fractional population dierence between the states involved in
the wave mixing; ,J : Raman shift; ,J : linewidth; /: dierential Raman cross section.
Population fractions of rovibrational states are given by the Boltzmann equation:
N,J =
G()
g J (2J + 1) kT F (J)
e vib e kTrot
Qvib Qrot
with g J : nuclear spin multiplicity; Qi : partition function. This exponential dependence of N,J on T
gives rovibrational spectroscopy its sensitivity to temperature.
Thermometry by coherent anti-Stokes Raman spectroscopy
289
Figure 14.2: Phase matching schemes.
The anti-Stokes wave is generated eciently in the direction of phase matching, that is k =
2kP kS kAS 0, or more generally
kAS = 2kP kS
The nonlinear wave mixing is a vector process, and the laser-like anti-Stokes signal leaves the diagnostic volume in a prescribed direction that depends on the vector angles of the pump and Stokes
beams (Fig. 14.2). The momentum conservation equation has to be fullled ensuring ecient coherent growth of the CARS wave. Note that once the pump laser frequency has been chosen, the Raman
shift of the probed molecule dictates the lengths of the wave vectors fullling energy conservation.
The simplest form of phase matching is the collinear approach but since the spatial resolution is
too low, the planar BOXCARS approach was developed. To spatially separate all output beams folded
BOXCARS was introduced.
Phase-matching can be bothersome and lead to loss of signal if not fullled, but has the ad-vantage
that all of the signal eld can be easily collected, as contrasted with Raman scattering. Suppression
of background radiation from ame light which is scattered in all directions is much better than in the
spontaneous Raman case. Since the CARS signal is blue shifted com-pared to the laser beams used,
discrimination against induced uorescence is easy.
The transition energies (cm1 ) in the Q-branch equal (neglecting higher order terms):
,J = e 2e xe ( + 1) e J (J + 1)
with e : vibrational wavenumber; xe : anharmonicity constant; e : rotation-vibration interaction constant.
Since in most (air-fed!) ames nitrogen is present, CARS uses N2 as the indicator molecule.
At room temperature, the N2 molecules are in the lowest vibrational level of the electronic ground
state and will possess a few quanta of rotational energy. In Fig. 14.2, the measured CARS spectrum
shows transitions for = 0 = 1 with low J numbers (cold band). At 1000 K, not only
higher J levels are occupied and, hence, show up in the spectrum, but also the = 1 vibrational level
becomes occupied and transitions for = 1 = 2 become visible as well (hot band). At ame
temperature, transitions for both vibrational levels with a lot of J numbers are observed.
The cold and hot bands are the relevant parts of the N2 Q-branch spectrum for thermometry, and
are conned to a small spectral region (2270 2330 cm1 ). They can be generated in a single laser
290
Figure 14.3: Normalized CARS spectra of N2 measured at 300, 1000 and 2200 K.
pulse of 10 ns from a broadband Stokes laser. Spectral separation is then achieved by using a spectrometer and detection is made by a CCD detector. This allows single-shot instantaneous temperature
measurements in non-stationary media, e.g. turbulent ames and transient ows.
14.3
Thermometry
Variation of temperature (and concentration) alters the shape of CARS spectra and analysis of this
shape underlies CARS thermometry. In practice, theoretically calculated CARS spectra at dierent
temperatures are t to the measured spectra, and the temperature producing the best t is taken as the
actual temperature. If the rovibrational quantum states are populated accord-ing to the Boltzmann law,
a rovibrational temperature is obtained. In most combustion systems this temperature is equal to the
translational temperature. Examples of CARS spectra are given in Fig. 14.2. The 2200 K spectrum
has been measured in a methane/air at ame stabilized on a McKenna burner. In such ames and
with the experimental setup described below, a single-shot imprecision of 20 K can be attained [3].
The 95% condence interval is estimated to be 20 K.
The CARS intensity strongly depends on temperature: the intensity at 300 K is about 10.000
higher than the intensity at 2500 K. The reason for this is twofold. In the rst place the densities
dier by a factor of 8. In the second place the width of the spectrum is smaller at low temperatures.
Remember that the CARS intensity scales with the square of density and population fraction. In
ames with steep temperature gradients this is a cause of inaccuracy. With only a small portion of the
probe volume in a cold region the spectrum is so much inuenced by the signal from this region that
an accurate probe volume averaged temperature is impossible to determine. The erroneous bias to low
temperatures is easily detected since no theoretical spectrum will properly t the measured one.
At atmospheric pressure, species spectra show closely spaced and partially overlapped Raman
transitions which are predominantly collision broadened. As the pressure (density) increases and
the lines broaden to the point of substantial overlap, rst constructive, coherent line mix-ing sets
in followed by collisional narrowing. The key to accurate spectral synthesis resides in the scaling
models used to predict the Raman line widths as a function of density, temperature, and rotational
quantum number. The modeling laws are semi-empirical and generally work best over some restricted
parameter range.
A clear drawback of CARS is that single-point temperature data are obtained. Mapping 2D temperature distributions will take a lot of time.
291
Figure 14.4: Experimental setup: BS, beam splitter; T, telescope; L, lens; BD, beam dump.
14.4
Experimental setups
A state-of-the-art CARS experimental setup is shown in Fig. 14.4. In principle a lot of pump lasers
can be used, but typically a frequency-doubled Nd-YAG laser is applied. This work-horse is injectionseeded by a diode laser (among others to get high spectral resolution). Its second harmonic at P = 532
nm has a pulse energy of 500 mJ at a 10 Hz repetition rate. The pulse duration is 10 ns, the beam
diameter is 10 mm, and the linewidth is smaller than 0.003 cm1 . Approximately 80% of the vertically
polarized laser output is used to pump a modeless dye laser. The dye laser contains a pre-amplier
with an interference lter for spectral ltering and tuning, and two longitudinally pumped ampliers,
yielding a 150 cm1 FWHM broadband prole at 605 608 nm for Rhodamine 101 in methanol. A
telescope is applied to expand the beam diameter and to control the position of the beam waist in the
probe volume. The remaining 20% of the pump laser travels along a delay line (to compensate for
the long beam path in the dye laser) and is split into two beams of equal intensity. Typical energies
used to generate the CARS beam are 40 mJ for the 532 nm pump and probe beam and 15 mJ for the
broadband Stokes beam. Application of the here described laser combination guarantees low noise
level in the spectra. An aplanat focuses the beams into a cigar-shaped probe volume measured to
be 0.7 mm long and 35 m in diameter. In the focal point high irradiances of 1012 W/cm2 are present,
and one has to take care of optical breakdown and stimulated Raman scattering.
Beyond the probe volume the three strong beams are dumped. Although a laser beam, the generated CARS beam has too low intensity to be visible by eye. The beam at 473 475 nm is collimated,
spectrally dispersed in a high resolution spectrometer, and recorded on a back-illuminated CCD with
17-bit dynamic range. As shown in Figure 3, spectra are obtained in which the rotational structure is
resolved. These spectra and theoretical ones are used to determine the instrument broadening function, in consequence of the applied spectrometer and detector. The shape of the broadband Stokes
prole is Gaussian, and its actual excitation prole is determined from a nonresonant CARS spectrum
in pure methane or argon. By blocking the Stokes beam the baseline level can be measured. Theoretical spectra are calculated using dedicated software, including CARSFT [4] and Da-capo [5, 6]. They
are convoluted with the instrument function, multiplied by the excitation prole, and stored in a data
base, in a range of for example 250 3000 K with steps of 25 K. Measured N2 CARS spectra are
baseline subtracted, and tted with the library spectra. In the least-squares t, temperature, concentration, frequency shifting and vertical stretching are varying parameters, and interpolation is used to
get accurate spectra for intermediate temperatures.
Spectroscopic parameters, needed to calculate reliable spectra, are accurately known for N2 . The
same is true for O2 . O2 CARS spectra can be measured using Rhodamine B emitting its broadband
prole at 577-581 nm. Also H2 and CO2 thermometry can be used.
292
CARS measurements of species concentration are performed in either of two ways: from the
intensity of the spectrally integrated resonant signal, or from the shape of the spectrum. In the former
approach, the measurements must be referenced appropriately, which is not straightforward. The latter
approach can be employed, in certain concentration ranges, when the resonant signal interferes with
the nonresonant background electronic contributions. CARS is limited to measurements of major
species (N2 , O2 , H2 , CO, . . . ), i.e., those occurring in abundances greater than approximately 0.1%.
A disadvantage of rovibrational CARS is that dierent experimental setups are required to probe
Raman resonances of dierent species. However, smart pump schemes can be applied to simultaneously measure spectra of dierent species. One such scheme is dual-pump CARS [7], in which the
probe beam at 532 nm is replaced by a beam from a tunable narrow band dye laser. As an example,
a 532 nm pump beam and a 607 nm Stokes beam create N2 resonances from which a 554 nm probe
beam scatters, whereas, using the same beams, a 554 nm pump beam and a 607 nm Stokes beam
create O2 resonances from which a 532 nm probe beam scatters. Both CARS signals show up in the
same spectral window. Simultaneous excitation has been realized for combinations including N2 -O2 ,
N2 -H2 , N2 -CO2 , N2 -CH4 , O2 -CO, and even N2 -O2 -H2 -C2 H4 -CO-CO2 , resulting in relative species
concentrations. Since concentration determination by CARS is not a main objective of this course, it
will not be discussed here in further detail.
Besides rovibrational CARS, another popular setup is dual-broadband pure-rotational (DBPR)
CARS [8]. Here, a broadband dye laser beam is split into two beams. Dierent frequency combinations within the broadband proles excite rotational Raman resonances from which a 532 nm
probe beam scatters. Since rotational resonances have small Raman shifts, a DBPR-CARS spectrum
contains spectral lines from all molecules relevant for combustion (if the signal is strong enough).
Rotational CARS is of interest due to its thermometric accuracy at low temperatures ( 1000 K) and
avoidance of collisional narrowing until high pressures be-cause of its well separated transitions.
14.5
Outlook
The recent availability of o-the-shelf picosecond (ps) and femtosecond (fs) lasers oers interesting
possibilities [9]. A drawback of the ns (nanosecond) laser-based approach is the interference of the
nonresonant background signal with the resonant signal, in particular in hydrocarbon-rich environments. This interference will occur only when the pump, Stokes, and probe beams are overlapped
temporally. In the ps regime, it is possible to delay the probe beam temporally with respect to the
pump and Stokes beams in an attempt to suppress the nonresonant contribution to the CARS signal. It
was shown that with 100 ps beams and 100 ps delay time the nonresonant signal could be reduced
by more than three orders of magnitude, whereas the resonant signal was reduced by only a factor of
three. A fundamental problem regarding possible distortion of the time-delayed spectrum must be addressed before ps-CARS can be used reliably for temperature measurements. The various transitions
that contribute to the Raman resonances and constitute the ps-CARS spectrum relax at rates that can
dier from transition to transition.
Measurements in the ns and ps regimes are generally performed at 10 Hz. This frequency is not
sucient for addressing the evolution of various instability modes and their interactions related to turbulent reacting ows. Furthermore, interpretation of measured signals requires a detailed understanding of the eects of molecular collisions. Since no unique approach has been developed for modeling
the eects of collisions, a semi-empirical approach is generally required. An optimal measurement
technique for use in practical combustors would not only be independent of the local eects of molecular collisions but also provide high-bandwidth data acquisition capability. In time-resolved fs-CARS,
293
measurements can be performed on a time scale that is shorter than the characteristic collision time
of the probed environment, for pressures up to 2 MPa. Recently [10], single-shot temperature measurements at a data rate of 1 kHz employing fs-CARS have been demonstrated.
References
[1] A.C. Eckbreth, Laser Diagnostics for Combustion Temperature and Species, Gordon and Breach,
Amsterdam, 1996.
[2] F.S. Moya, S.A.J. Druet, and J.P.E. Taran, Gas Spectroscopy and Temperature Measurement by
Coherent Raman Anti-Stokes Scattering, Opt. Comm. 13 (1975) 169-174.
[3] E.H. van Veen, and D. Roekaerts, Thermometry for turbulent ames by coherent anti-Stokes
Raman spectroscopy with simultaneous referencing to the modeless excitation prole, Appl. Opt.
44 (2005) 6995-7004.
[4] G. Clark, and R.L.Farrow, The CARSFT code. User and programmer information, Sandia National Laboratories, Livermore, CA. (1990).
[5] D. Brggemann, and S. Heshe, DACAPO. CARS software, version 94.A, ITLR, University of
Stuttgart (1994).
[6] E.H. van Veen, and D. Roekaerts, On the accuracy of temperature measurements in turbulent
jet diusion ames by coherent anti-Stokes Raman spectroscopy, Combust. Sci. and Tech., 175
(2003), 1-22.
[7] R.P. Lucht, Three-laser coherent anti-Stokes Raman scattering measurements of two species,
Opt. Lett. 12 (1987) 78-80.
[8] M. Ald n, P.-E. Bengtsson, and H. Edner, Rotational CARS generation through a multiple foure
color interaction, Appl. Opt. 25 (1986) 4493-4500.
[9] S. Roy, J.R. Gord, and A.K. Patnaik, Recent advances in coherent anti-Stokes Raman scattering
spectroscopy: fundamental developments and applications in reacting ows, Progress in Energy
and Combustion Science 36 (2010) 280-306.
[10] S. Roy, W.D. Kulatilaka, D.R. Richardson, R.P. Lucht, and J.R. Gord, Gas-phase single-shot
thermometry at 1 kHz using fs-CARS spectroscopy, Opt. Lett. 34 (2009) 3857-3859.
294
Chapter 15
Application of laser diagnostics

Th.H. van der Meer
15.1
Introduction
One of the biggest challenges in turbulent combustion is its accurate modeling. The modeling should
help to organize a combustion system in such a way that the combustion process is most ecient, the
heat transfer is optimal, emission of pollutants are minimal, and that product quality is maximal. Since
this is not an easy task in which many model assumptions have to be made there is a need for comparisons of model predictions with experimental data. These experiments should be done under well
dened conditions, leaving as little room as possible for the modeler to tune the models into the direction of the experimental data. Experiments on an industrial scale for this reason are not very suitable
for detailed model validations. Over the years experimental databases have been gathered from experiments on laboratory ames using dierent gaseous fuels in dierent burner congurations. These
ames have been discussed in an international forum and are now well accepted as standard ames
for model comparisons. One of these ames is the Delft piloted diusion ame, which is the topic of
this contribution. Here results from various experimental studies are gathered (see references[1][7]).
15.2
The Delft piloted diusion ame burner
The Delft piloted diusion ame burner was used to produce a laboratory-scale axisymmetric turbulent diusion ame. The burner is described by Peeters et al. [1] and in full detail in the theses of
Stroomer [3] and de Vries [2]. The burner consists of a central fuel jet, surrounded by two concentric
co-ows of air. Figure 15.1 shows the burner head.
The fuel jet nozzle is 6 mm in diameter. It is separated from the primary air stream by a 4.5 mm
wide rim. Twelve small pilot ames, necessary for the stabilization of the ame on the burner, are
issued from holes located on this rim. The outer diameter of the primary air annulus is 45 mm. The
length of the burner is 100 cm. In the rst 94 cm, the inner diameter of the primary air annulus is
30 mm, decreasing to 15 mm in the nal 6 cm. This gives rise to a small negative radial velocity
component of the primary air in the exit plane of the annulus. The initial diameter of the central fuel
pipe is 8 mm. A pilot ame insert placed in the exit of the fuel pipe causes a decrease in diameter to
6 mm starting at a position 16 mm upstream of the nozzle exit. Although the decrease in diameter is
gradual, with a 7 angle, the ow leaving the fuel pipe cannot be considered to be fully developed.
Experiments were performed for dierent sets of inlet boundary conditions. In this contribution
we will restrict ourselves to two ames, denoted Flame III and IV. Figure 15.1 shows photographs
of these ames. Flame III is considered as a reference ame or base case. The fuel jet velocity
was 21.9 m/s and at room temperature (Re = 9700). The primary air ow velocity was 4.4 m/s, also
295
296
Chapter 15 Th.H. van der Meer
Figure 15.1: Top and side view and photograph of the burner head, dimensions are in mm. At the right
photographs of Flame II and Flame IV.
at room temperature (Re = 8800). In Flame IV the ow rate of the primary air stream was doubled
(Uannulus = 8.0 m/s, Re = 16000) to study the eect of dierent turbulent mixing rates. For both
ames, the secondary air stream was kept at a velocity of 0.3 m/s. The air ow was produced by
positioning the burner in a wind tunnel with an exit cross-section of 254 254 mm.
The pilot ames were fed with a premixed acetylene/hydrogen/air mixture with an equivalence
ratio of 1.4. The C/H ratio was the same as that of the natural gas. The heat release in the pilot
ames was about 1% of the total thermal power of Flame III. A small recirculation zone located
directly above the narrow rim provided a second mechanism for ame stabilization.
A large number of measurements using dierent techniques have been performed in the ames
from this burner. The techniques that have been used are Laser Doppler velocimetry (LDA) for velocity measurements, Laser Induced Fluorescence (LIF) for the measurements of radical species concentrations, Coherent anti-Stokes Raman spectroscopy (CARS) for temperature measurements. These
experiments were performed in the Thermouids group of Delft University of Technology. Next
in cooperation with Sandia National Laboratories combined Rayleigh-Raman-LIF for simulataneous
measurements of species concentrations and temperature were done. Finally time resolved PLIF measurements of OH were done in cooperation with the Division of Combustion Physics of Lund Institute
of Technology. In the following measurements from these studies will be discussed.
15.3
15.3.1
Experimental results
LDA experiments
Measurements of the uctuating velocity component at a position of 50 mm. above the burner exit at
two positions in Flame IV are given in gure 15.2 together with results from temperature measurements and measurements from OH-concentrations. Figure 15.3 shows the same variables at a position
of 250 mm. above the burner exit.
297
Figure 15.2: Radial proles of velocity uctuation, OH-concentration and temperature at axial location of
50 mm
Figure 15.3: Radial proles of velocity uctuations, OH-concentration and temperature at axial location of
250 mm.
Figure 15.2 shows symmetric proles of urms , cOH and T with peaks is the region where the
natural gas from the inner tube mixes with the air from the annulus. The position of the ame front
can be expected at the stoichiometric value of the mixture fraction, being 0.06. From gure 15.2 it is
clear that the position where the OH-concentration and the temperature have their maximum, does not
coincide with the position where the uctuating velocity is maximal. This position of max. cOH and T
is on the air side of the shear layer between natural gas and air from the annulus. This is in agreement
with the position where the mixture fraction has a the stoichiometric value. From gure 15.3 this
dierence in position of the maxima is not present anymore. The turbulent mixing process now has
caused much broader peaks at this location.
From autocorrelation functions of axial velocity uctuations at several locations turbulent integral
time scales were found in the ames. Taylors hypothesis was used to nd estimates of turbulent integral length scales from these time scales. In such a way at the centerline of ame IV at axial distance
of 250 mm. an integral length scale of 7.5 mm. was found. Also from 1D LIF measurements of OHconcentrations and NO-concentrations turbulent length scales of OH-structures and NO-structures
could be derived. This led to an integral scale for NO of 3.1 mm. and an integral scale of OH equal
to 1.9 mm. At the radial location where the turbulence intensity has its maximum also estimates of
these scales were found: a velocity scale of 8.5 mm., from NO measurements also 8.5 mm, from
OH-measurements 1.5 mm. The integral OH scales are much smaller than the velocity and NO scales.
This is because OH is strongly inuenced by chemical reactions. OH exists in very narrow regions
and the chemical time scale of the reactions in which OH takes part must be smaller than the integral
298
Figure 15.4: The Raman-Rayleigh-LIF system
velocity scale. This may lead to the smaller integral OH-scales, which will be conrmed later by time
resolved OH measurements.
15.3.2
Rayleigh-Raman-LIF measurements
The experimental set up for simultaneous temperature and species concentration measurements is
shown in gure 15.4. This set-up is described in detail elsewhere [8][12]. The set up consisted
of four separate laser systems. One Nd:YAG laser for the Raman-Rayleigh measurements and three
separate Nd:YAG/dye laser systems for the LIF measurements of OH, CO and NO. Because all lasers
were red almost simultaneous quantitative corrections were possible for collisional quenching of the
LIF signals. Interferences between the dierent systems were avoided by the application of a timing
delay of 100 ns, small compared to typical chemical and turbulent length scales. The vibrational
Raman signals of CO2 , O2 , CO, N2 , CH4 , H2 O and H2 were recorded using PMT tubes aligned along
the exit plane of a polychrometer. Three additional PMTs were positioned in between the Raman
channels to mionitor the uorescence background. The probe volume length was estimated to be
0.75 mm.
15.3.3
N2 , CO2 and O2
The interpretation of the Raman signals was complicated by interferences with Raman scattering from
other species and with non-resonnant uorescence from heavy hydro-carbons such as PAHs and other
soot precursors. The cross-talk with other species was taken into account by a calibration procedure
involving the recording of Raman signals over a wide range of temperatures, gas compositions and
299
Figure 15.5: Scatterplot of the N2 Raman signal versus the uorescence interference signal without (left) en
with (right) interference correction (Flame III, x = 150 mm, r = 8 mm).
Figure 15.6: Scatterplot of the CO2 Raman signal versus the uorescence interference signal without interference correction at x = 150 mm. Flame III (left), Flame IV (right).
concentrations. The broadband uorescence interferences from the heavy hydrocarbons were monitored by PMTs positioned at spectral locations in between the Raman channels. The PMT at the
615 nm channel provided the best signal-to-noise ratio and was used in the correction procedure. The
correction procedure for uorescence intereferences is illustrated in the scatter plots in Figure 15.5.
The uncorrected N2 Raman signal in the left graph shows a positive correlation with the interference
channel. To the right, the uorescence contributions to the single-shot N2 concentration have been
corrected and the resulting N2 concentration shows no correlation with the uorescence signal.
For the species O2 and CO2 the interference correction cannot be performed in the above manner as for N2 . Figure 15.6 shows the scatterplots of CO2 versus uorescence interference signal at a
typical position in Flames II and IV. The scatter plot for ame III shows two distinct branches, apparently corresponding to two dierent uorescence interference modes. In ame IV, the interferences
are weaker and, more importantly, only the lower branch is observed. Clearly, the structure of the
uorescence interferences in ame III is such that it is not possible to correct for it using a single
linear correlation with the monitor channel. It was found that the upper branch corresponds to uorescence from richer mixtures at lower temperatures. However, no useful correlation between this
additional uorescence and the interference contribution to the CO2 Raman signal could be identied.
In the processing of the data collected in ame III, it is therefore preferred not to use the CO2 Raman
300
Figure 15.7: Comparison of centerline dry CO2 (left) and O2 (right) mole fractionproles for ame III.
signals from points with instantaneous signals on the 615 nm uorescence monitor channel above
a certain threshold. For these points, the CO2 concentration is obtained from an indirect procedure
using the measured Rayleigh temperature and data from a laminar amelet calculation instead. In
this procedure, it is assumed that the thermochemical composition in the ame is close to that in a
laminar amelet with strain rate a = 100 s1 . A progress variable based on the Rayleigh temperature, introduced below, is used to describe small deviations from this composition. Simulations show
that ame III exhibits a wide range of strain rates. However, realistic CO2 concentrations can be obtained with a procedure based on a single representative strain rate. The same procedure was used
for the determination of O2 concentrations. Obviously, the resulting CO2 and O2 concentrations are
strongly correlated with temperature. As such, they do not provide independent information on the
interaction between the turbulent ow eld and the chemistry. The main purpose of the procedure is
to obtain CO2 and O2 concentrations that are adequate for the evaluation of the mixture fraction and
the eective Rayleigh cross- section. For further analysis of the limitations of the above method for
determining CO2 and O2 concentrations the mean values were compared to probe measurements. For
Flame III on the centerline at x = 150 mm. this comparison is shown in Figure 15.7.
On the centerline, the CO2 and O2 Raman signals suer from substantial contributions of uorescence interferences. Consequently, the CO2 and O2 concentrations are overestimated if calculated
directly from the Raman signals. The values from the laminar amelet procedure show a reasonable
agreement with the probe data. For the radial traverse at x = 150 mm, similar results are obtained.
These results indicate that the CO2 and O2 concentrations from the laminar amelet procedure are
more reliable than those derived directly from the Raman signals.
15.3.4
OH, NO and CO
Here some results of the OH, NO and CO measurements are presented. The OH and NO LIF signals
are calibrated by comparing them to signals from ames with well-known concentrations. For OH, a
premixed CH4 air ame from a Hencken burner with an equivalence ratio of 0.95 is used. The nominal
OH concentration in this ame is 1.44 105 moles.dm3 , determined by a laser absorption measurement. This concentration is close to the non-adiabatic equilibrium concentration at the reference
ame equivalence ratio and at a temperature 50 K below the equilibrium temperature. The NO system
is calibrated using a premixed CH4 -N2 -O2 ame ( = 0.72), stabilized on a McKenna burner. By
replacing part of the N2 by NO, ames with dierent concentrations of NO can be produced. Laminar
ame calculations show that the doped NO is conserved through the ame front. Thus, the signals
301
Figure 15.8: Mean resonant and non-resonant OH uorescence signals (left) and scatterplot of uncorrected OH
concentration (molecules cm3 ) versus mixture fraction (middle) at x = 150 mm in ame III. To the right, the
eect of interference correction on the mean OH concentration (molecules cm3 ) at x = 150 mm in ame III.
recorded from ames with dierent known doped NO levels can be used to construct a calibration.
The single shot temperature and composition measurements are used to correct for:
the dierence between the population fractions at the reference temperature in the calibration
ame and the temperature in the probe volume,
the collisional quenching rate in the calibration ame relative to the collisional quenching rate
in the probe volume,
the dierent spectral overlaps between the laser and the selected transition at the reference
temperature in the calibration ame and at the probe volume.
Although the LIF signals are stronger than the Raman signals, they still are aected by interferences
from heavy hydrocarbons. This is mainly because of the wider spectral intervals over which the signals
were collected. The interference contribution to a LIF signal was estimated from the o-resonance
signal, where the dye laser system was tuned to a wavelength where none of the probed molecules
were excited. As the uorescence interferences from heavy hydrocarbon species are generally broadband, the o-resonance signal is considered to be a reliable estimate for the interference contribution
to the LIF signal. The o-resonant signals cannot be measured simultaneously with the resonant
signals. Consequently, they cannot be used to subtract the interference contributions from the LIF
signals on a single-shot basis. However, the o-resonant measurements provide valuable insights into
the distribution of the interferences in both physical and composition space.
The left graph in gure 15.8 presents the mean resonant and non-resonant OH signals measured at
x = 150 mm in ame III. It is seen that near the centerline, for r < 5 mm and where > 0.4, nearly the
entire recorded LIF signal originates from interferences. At the positions where the maximal resonant
OH signals are measured, the non-resonant contribution is only about 2%. The middle graph shows
the distribution in mixture fraction space of the OH concentration calculated from the uncorrected
resonant signal. Clearly, the non-zero OH concentrations for > 0.4 stem from interferences: the
OH is expected to be conned to a narrow zone located around stoichiometric mixture fraction. The
non-resonant contributions to the measured OH concentrations can therefore be removed by setting
the OH concentrations to zero for mixture fraction values above a certain threshold. The threshold
value employed in this work is 0.20. This value is not critical: threshold values of 0.10 and 0.30 give
302
Figure 15.9: Mean resonant and non-resonant NO uorescence signals (left) and scatterplot of non-resonant
NO signal versus the uorescence interference channel (right) at x = 150 mm in ame III.
Figure 15.10: Mean resonant and non-resonant CO uorescence

signals at x = 150 mm in ame III
nearly identical results for the mean OH concentration. The unphysical non-zero OH levels found
near the centerline position in the uncorrected prole are hereby eliminated. In the zones with high
OH concentrations around r = 14 mm, the non-resonant contributions are negligible and the corrected
and uncorrected proles coincide.
For the case of NO, a dierent approach was employed to suppress non-resonant interferences.
The left graph in gure 15.9 shows the mean resonant and non-resonant signals at x = 150 mm in
ame III. The ratio of the non-resonant and resonant signals is found to be approximately constant
across the prole. Scatter plots show that the structure of the NO o-resonance signal in mixture
fraction space does not allow for a correction analogous to that used for OH. However, the o-resonant
signal exhibits a strong correlation with the uorescence interference channel monitored at 615 nm
(gure 15.9 to the right). This correlation is used to subtract the interference contributions from the
resonant signal before the NO concentration is calculated.
Figure 15.10 presents the mean resonant and non-resonant CO signals at x = 150 mm in ame III.
The non-resonant contribution is some 8% of the total signal. The CO non-resonant signal is found to
exhibit a correlation with the interference channel, although not as strong as in the case of NO. This
correlation is incorporated in the response matrix used in the inverse Raman signal problem. Using
this approach, the non-resonant contributions are reduced to about 5%. In ame IV, the absolute nonresonant OH signals at x = 150 mm are found to be typically four times smaller than those in ame III.
The absolute OH concentrations are somewhat smaller as well. The non-resonant contribution near
the centerline is about 50%. The non-resonant NO and CO signals at x = 150 mm in ame IV are
303
Figure 15.11: Comparison of measured mean and

RMS of temperature at x = 100 mm (bottom)
and 200 mm (top) in ame III. Mean temperatures of CARS (red), Reynolds-averaged RamanRayleigh (blue) and Favre-averaged RamanRayleigh (green) measurements are indicated with
solid symbols. Open symbols give temperature
standard deviations.
typically six and three times smaller, respectively, than those at the same position in ame III.
15.3.5
Temperature
Figure 15.11 presents proles of the mean and standard deviation of temperature from the RamanRayleigh measurements together with the CARS data for two axial locations in ame III (x = 100 mm
and x = 200 mm). As the instantaneous density is available in the Raman- Rayleigh measurements,
it is straightforward to compute both the Favre- and Reynolds-averaged temperature for these data.
The density weighting in the Favre averaging procedure gives rise to lower mean temperatures. The
dierence between the Reynolds and Favre mean temperature varies from about 15 K at positions near
the centerline to some 350 K in the outer anks of the proles. The Favre peak mean temperatures
are typically 120 K lower than the Reynolds values. The maximum and centerline values of the
temperature standard deviations are similar. The Favre-averaged proles show a faster decay in the
ank region, consistent with the shape of the mean temperature prole. It is unclear whether the CARS
measurements yield conventional (Reynolds) or density-weighted (Favre) averages. The CARS mean
temperatures in gure 15.11 are nearly identical to the Reynolds-averaged Raman-Rayleigh results,
especially in the ank regions of the ame. At the centerline, where the dierences between Reynoldsand Favre-averaged temperatures are relatively small, the CARS and Raman-Rayleigh means agree
well and are within the experimental uncertainty. The CARS temperature standard deviations are
closer to the Reynolds-averaged Raman-Rayleigh values as well. Comparison of the dierent datasets
at other axial locations yields similar results.
Figure 15.12 shows temperature PDFs measured at four radial positions at x = 150 mm in
ame III. The graphs compare the CARS PDFs with the Reynolds and Favre PDFs obtained from
the Raman-Rayleigh measurements. The CARS and Raman-Rayleigh PDFs are constructed from
5000 and 4000 points, respectively. For the relatively narrow, monomodal PDF found at the centerline, the Reynolds and Favre PDFs are very similar (gure a). Except for a small dierence in standard
deviation, they agree well with the CARS result. At r = 7.5 mm (gure b), the PDFs are somewhat
304
Figure 15.12: Comparison of measured temperature PDFs for dierent radial positions at x = 150 mm in
ame III. Favre-averaged (green) and Reynolds-averaged (blue) PDFss constructed from the Raman-Rayleigh
measurements. CARS results from Zong et al. Positions: r = 0.0 mm (a), r = 7.5 mm (b), r = 11.5 mm (c) and
r = 15.5 mm (d).
Figure 15.13: Scatterplot of the measured OH

concentration (molecules.cm3) at x = 150 mm.
The lines represent the Favre conditional average (red), a laminar amelet calculation with a =
100 s1 (magenta) and and chemical-equilibrium
calculation (blue).
wider. For the bimodal PDF at r = 11.5 mm, gure c, the Favre and Reynolds PDF shapes dier
substantially. The density weighting in the Favre PDF produces a larger probability at lower temperatures. At this position, the CARS PDF is closer to the Reynolds than to the Favre result. Figure
d shows the PDFs measured in the outer ank of the temperature prole (r = 15.5 mm). Although
the three results are qualitatively similar, the relative weights of the room temperature peak and the
high-temperature tail are dierent for the Favre PDF on one side and the Reynolds and CARS PDFs
on the other side.
Figure 15.13 shows typical COH - scatterplot data obtained from the Raman-Rayleigh-LIF experiments in ame III, together with the proles from a 100 s1 amelet and a chemical-equilibrium
calculation. Comparison of the scatterplot data to the equilibrium prole clearly shows that OH is
present in super-equilibrium concentrations in the upstream part of the ame. The OH levels predicted
by the equilibrium model are some 3 to 5 times lower than the measured values. The occurrence of
super-equilibrium OH concentrations is a well-known phenomenon in the type of jet ame considered
305
here. Figure 15.13 shows that super-equilibrium OH levels are similar to those observed in moderately
strained amelets.
15.3.6
Time resolved PLIF of OH
Although much insight into the natural gas ames has been gained from the measurements discussed
above, time resolved measurements are needed to visualize the evolution of turbulent reactive structures. With these measurements a better understanding of turbulence chemistry interactions can be
obtained. Time resolved measurements also add to a more complete database.
In this paragraph, we present time-resolved OH PLIF measurements, at repetition rates of the
same order as the characteristic frequencies of the ow, in turbulent diusion ames from the Delft
burner with Re number in the range of 7000 to 16000. We show how turbulent structures interact with
the reaction zone in the turbulent diusion ames and track their development in a lm like manner.
For details on the laser system the reader is referred to [13]. Briey, the laser consists of 4
individual Nd:YAG oscillator/amplier units (B.M. Industries, France). The individual channels are
each equipped with a double pulse option (DPO) where the Pockels cell is switched twice during the
duration of the ashlamp pulses. In this way two pulses can be extracted per cavity with selectable
delays ranging from 25 to 145 s. Thus a total of 8 pulses can be obtained from the system
propagating as a single beam. Pulse energies at 532 nm are around 600 mJ per pulse in single pulse
mode (total of four pulses) or around 270 mJ per pulse in DPO mode. For the experiments presented
here, the laser was mostly run with a total of four pulses per sequence thus aording higher pulse
energies and better signal to noise ratios.
A single, frequency doubled, dye laser system (Continuum) operating with Rhodamine 590 dye
containing methanol solutions was pumped with this laser pulse train. Due to thermal loading eects
and dye degradation in the dye laser the resulting beam prole and intensity was degraded from one
pulse to the next within a sequence. This dye memory eect made shot-to-shot referencing of the
beam prole essential (see subsequent paragraphs for details).
Typically around 810 mJ per pulse around 283 nm were obtained with a linewidth of 0.3 cm1 .
The wavelength was tuned to the Q1 (8) transition in the v = 1 v = 0 band of the A2 + X 2
system of OH. For all experiments OH uorescence in the 1 1 and 1 0 bands near 310 nm
were detected by passage through bandpass interference lters. The light was spatially ltered and
focused into a sheet whose dimensions in the interaction region were approximately 50 mm in height
and 0.25 mm in thickness.
For the concentration measurements described, laser sheet proles were recorded online which
could be used to normalize PLIF data on a shot by shot basis. For this purpose, a reection of the laser
sheet was directed into a cell containing uorescing dye. The resulting LIF signals were imaged onto
the detector concurrently with the OH PLIF images which were subsequently divided by this prole.
The procedure is described in detail in [12].
A custom modied Imacon 468 framing camera (Hadland Photonics, UK, see [12] for details) was
used as a detector for the PLIF signals. The system features 8 independent intensied CCD detectors
(ICCD) with 384 576 pixel arrays and 8 bit dynamic resolution. To increase the overall gain of the
system and to make it UV sensitive an additional three-stage intensier module was attached to the
optical input of the camera prior to the prism beam- splitter.
Measurements were performed at eight heights from 0 to 395 mm above the nozzle for each
ame. Dierent time separation between events in a recorded sequence at 125 s, 250 s, 500 s,
1 ms and 2 ms were varied to track the dierent temporal scales of the turbulent eddies. Several
306
Figure 15.14: OH island formation: Time sequence OH PLIF images of Flame III show the formation of an
OH island indicated by the arrow. Image center height is 370 mm from nozzle.
Figure 15.15: Flame vortices movement: Time sequence of OH PLIF images shows a ame vortices movement.
Images are of ame III, center height of upper images is 370 mm from nozzle. Lower images correspond to the
indicated box in the upper images.
representative images showing the time development of the turbulence-ame interaction are presented
in gures 15.1415.17.
Figure 15.14 shows OH island formation in ame III at the height of 345 to 395 mm downstream
from the nozzle. In the remainder of this discussion we will refer to center heights only, which is for
this case 370 mm. In image (a) (t=0 ms), a very low intensity OH layer (indicated by the arrow) is
wrinkled, most probably because of an air-sided anticlockwise turbulent eddy near the air/fuel shear
layer. The wrinkle develops into a ame tongue wrapped around the eddy and the tip of the tongue
connects with the upper thick OH area in the next 1ms. The tip of the ame tongue becomes thick
and strong due to both the strains of the eddy and the connection with the upper thick ame area.
At t = 2 ms, the ame tongue highly curves and almost touches itself. It stretches the upper thick
ame area and makes it thin and weak. Finally, at t = 3 ms in image (d), the ame tongue touches
itself, resulting in a ame island separating from the main body. The ame island becomes smaller by
consuming the oxidizer inside the island, while it continues moving upwards and rotating. The time
scale of the formation of an OH island is of the order of several ms. The length scale of the turbulent
eddy, which causes the formation of the ame island, is about 10 mm.
At the same axial position in the same ame another interesting event is illustrated in Figure 15.15.
Further details are presented in the lower enlarged images (a) to (d). At t = 0 ms, an upstream
ame tail caused by turbulent movement is seen in image (a) (indicated by arrow 1). After 1 ms it
has become thinner, especially in the base position (indicated by arrow 2). It could be caused by
307
Figure 15.16: Flame extinction: Time sequence OH PLIF images shows occurrence of a local extinction
(indicated by arrow 1). Images are of ame III, at the image center height of 270 mm from nozzle.
Figure 15.17: Extinction and re-ignition: Time sequence OH PLIF images from ame IV at a center height of
170 mm from nozzle. A local extinction is indicated by arrow 2. Arrow 1 and 3 indicate ame reignition.
a signicant strain due to an anticlockwise vortex-induced rotational motion. Then, at t = 2 ms, the
ame tail separates from the main ame layer. It follows the anticlockwise rotational movement while
moving up and consuming itself. In the last image at 3 ms. we see OH in the complete structure, which
means that gas and air were trapped inside the structure and reactions have taken place. Meanwhile
the main ame zone becomes thick too. It could be predicted that at the next moment the separated
ame tail would connect with the main ame and the topology changes into one thick ame zone.
The scale of the ame vortices, estimated from the images, is 5mm in diameter.
An occurrence of local extinction indicated by arrow 1 can be seen clearly in Figure 15.16. This
is caused by two turbulent eddies. A thick curved OH layer (indicated by arrow 2) curves and rotates
with time. Following this structure 2 in time shows that it rotates clockwise. The compressive strain
induced by the eddy thickens the ame layer and the extensive strain thins it. Such a mechanism
has been previously postulated to explain the oscillating ame layer thickness in jet ames . Another
wrinkle occurs in the lower OH layer indicated by arrow 3 in gure 15.16. It could be caused by the
same reason as the upper structure (indicated by arrow 2). The OH layer between these two curved
ame layers (indicated by arrow 1) undergoes signicant extensive strain caused by both eddies. The
result is that here the ame zone becomes thinner, and, nally, extinguishes due to the high strain (see
image (d)). The whole local extinction procedure occurs within several ms.
Figure 15.17 shows a time sequence of OH PLIF images in ame IV at a height of 170 mm
downstream. Very curved ame structures can be seen all over the image region. These ame structures show more and smaller wrinkling than those from ame III in gure 15.16. This is because the
Reynolds number of the air ow for ame IV is twice as high as that of ame III. A local extinction
is indicated by arrow 2. It is most possibly caused by either a high strain from the ame layer above
it or by strain coming from a turbulent eddy beside it. Arrow 1 and 3 highlight regions where a broken ame re-ignites. These re-ignited structures are thick pointing at a certain degree of premixing
during the period that the ame was locally extinguished. These results show the interaction between
combustion and small turbulent structures.
308
15.4
Conclusions
The application of the Raman technique in the undiluted natural-gas ames considered here proves to
be very challenging because of the high uorescence interference levels. The interference contributions to the recorded Raman signals are identied and subtracted using empirical correlations between
the Raman signals and the signals on the interference monitor channels. This procedure proves to be
adequate for most species. However, it is found that the empirical approach cannot be used to remove
the interference contributions to the CO Raman signal. This also holds for the CO2 and O2 signals in Flame III, the ame with the highest interference levels. The concentrations of these species
are determined using alternative approaches. Because of the uorescence interferences, the acquired
Raman-Rayleigh-LIF dataset has certain limitations, the most important of which is the absence of
independent CO2 and O2 measurements in Flame III. The obtained proles of mean temperature and
temperature uctuations agree very well with CARS temperature measurements obtained in Flame III.
In Flame IV, which shows substantial local extinction, small dierences in the boundary conditions
are expected to be important. The Raman-Rayleigh-LIF measurements of the OH concentration compare well with the existing semi-quantitative one-dimensional LIF data. The Raman-Rayleigh-LIF
experiments provide nearly all of the desired simultaneous measurements of temperature and major
species concentrations. The Raman-Rayleigh-LIF data therefore form a valuable and useful extension
of the existing database for the Delft piloted jet diusion ame burner. The lm-like results presented
here give a detailed view of the evolution of the ame front following turbulent eddies and vortex
motion. Several phenomena, which are characteristic for turbulent ames, were visualized in time
and analyzed qualitatively in this paper. A turbulent eddy causing the formation of a ame island was
clearly observed. The pattern of a ame vortex movementrotating, departing and attachingwas
catched in time. The secret of single ame extinction caused by turbulence was revealed from the time
sequent images. The extinction occurred when the ame zone was stretched too much by the turbulent eddies. Flame re-ignition was also visualized. In contrast with previous structural investigations
based on planar LIF measurements, time- resolved PLIF measurement results allow direct observation of both the spatial and the temporal eects of ame turbulence interaction. However, some of the
results are ambiguous to interpret due to the general limitation of 2D imaging techniques for the study
of essentially 3D phenomena. A 3D volumetric rendering technique is being developed in Lund Laser
Center to perform 3D OH concentration measurements.
References
[1] Peeters, T. W. J., Stroomer, P. P. J., De Vries, J. E., Roekaerts, D. J. E. M., and Hoogendoorn, C.
J. (1994) Comparative experimental and numerical investigation of a piloted turbulent naturalgas diusion ame, Twenty-Fifth Symposium (International) on Combustion/The Combustion
Institute, pp. 12411248.
[2] Vries, J. E. de (1994) Study on turbulent uctuations in diusion ames using laser induced
uorescence, Ph.D. thesis, Delft University of Technology.
[3] Stroomer, P. P. J., De Vries, J. E., Van Der Meer, Th. H. (1999), Eects of Small- and Large-scale
Structures in a Piloted Jet Diusion Flame, Flow, Turbulence and Combustion 62, pp. 5368
(1999).
309
[4] Mantzaras, J., and Van Der Meer, Th. H. (1997) Coherent AntiStokes Raman Spectroscopy measurements of temperature uctuations in turbulent natural gasfueled piloted jet diusion ames,
Combustion and Flame 110, pp. 3953
[5] Nooren, P. A., Wouters, H. A., Peeters, T. W. J., Roekaerts, D., Maas, U. and Schmidt, D.
Monte Carlo PDF modeling of a turbulent natural-gas diusion ame, Combustion, Theory and
Modeling 1, pp. 7996 (1997).
[6] Nooren, P. A., Versluis, M., Van Der Meer, Th. H., Barlow, R. S., and Frank, J. Raman-Rayleigh
and LIF measurements of temperature and species concentrations in the Delft piloted turbulent
jet diusion ame. Appl. Phys. B 71, pp. 95111 (2000).
[7] Ding, T. J., Van Der Meer, Th. H., Versluis, M., Golombok, M., Hult, J., Aald n, M., Kmamine
ski, C., Time-resolved PLIF measurement in turbulent diusion ames, Proc. Of the third Int.
Symposium on Turbulence, Heat and Mass Transfer, Nagoya, Japan, April 26, 2000, pp. 857
864.
[8] Nguyen, Q. V., Dibble, R. W., Carter, C. D., Fiechtner, G. J., and Barlow, R. S., RamanLIF
measurements of temperature, major species, OH, and NO in a methaneair Bunsen ame, Combustion and Flame 105, pp. 499510 (1996).
[9] Barlow, R. S. and Carter, C. D., Raman/Rayleigh/LIF measurements of nitric oxide formation in
turbulent hydrogen jet ames, Combustion and Flame, 97, pp. 261280 (1994).
[10] Carter, C. D. and Barlow, R. S., Simultaneous measurements of NO, OH, and the major species
in turbulent ames, Optics letters, 19(4), pp. 299301 (1994).
[11] Dally, B. B., Masri, A. R., Barlow, R. S., Fiechtner, G. J., and Fletcher, D. F., Measurements
of NO in turbulent nonpremixed ames stabilized on a blu body, in: TwentySixth Symposium
(International) on Combustion/The Combustion Institute, pp. 21912197 (1996) .
[12] Barlow, R. S., Dibble, R. W., Chen, J. Y. and Lucht, R. P., Eect of Damk hler number on supero
equilibrium OH concentration in turbulent non-premixed jet ames, Combustion and Flame 82,
pp. 235251 (1990).
[13] Kaminski, C. F., Hult, J., and Ald n, M., High repetition rate planar laser induced uorescence
e
of OH in a turbulent non-premixed ame, Applied Physics B 68, pp. 757760 (1999).
310
Part 5
Clean Combustion Concepts
311
Chapter 16
Preheated dilute combustion

D.J.E.M. Roekaerts and M.J. Tummers
313
314
Chapter 16 D.J.E.M. Roekaerts and M.J. Tummers
Chapter 17
Clean combustion concepts in furnaces

E.-S. Cho, B. Danon and W. de Jong
17.1
Introduction
Energy eciency and clean combustion are two main issues in fossil fuel utilization. A major industrial application of combustion is in process furnaces for heat generation and transfer to chemical
processes and products. Control of nitrogen oxides (NO x ) has been a major issue in designing combustion systems, since NO x plays a key role in acid rain formation and the generation of photochemical
smog [1]. Flameless Oxidation (FLOXT M ) [2], also known as High Temperature Air Combustion,
HiTAC [3, 4] or alternatively Moderate and Intensive Low oxygen Dilution, MILD combustion [5, 6]
is a promising combustion technology among various techniques [7] capable of accomplishing high
eciency accompanied with low emissions. It is based on delayed mixing of fuel and oxidizer and
high ue gas recirculation in the ame zone. High momentum injection of the separated fuel and air
ows entrain the ue gas through internal recirculation, thus diluting the oxygen concentration in the
combustion zone [8]. This leads to a more distributed heat release rate of the chemical energy, avoiding a high peak temperature (hot spot) and reducing the thermal formation of NO x [9]. Combined
with high preheat temperature of the combustion air, this combustion technique also achieves a high
eciency.
In earlier works, ameless oxidation has been studied in small ames to evaluate the characteristics of the turbulent ame temperature and structure using laser measurement techniques of Rayleigh
scattering and Laser Induced Fluorescence [10, 11]. Yang et al.[12] investigated a two-ame highcycle regenerative system numerically. Also, Radi and Blasiak [13] studied the characteristics of
ame interaction in this two-ame regenerative burning system experimentally varying the ring
mode (parallel, staggered and counter ring).
In all these studies, systems with one or two burner pairs are investigated. It is important to extend
these studies on congurations with more burners because in real industrial furnaces most often many
burners are present. Interaction between jets from dierent burners and the relative position of ring
and regenerating burners may play an important role.
At Delft University of Technology, a 3x100 kWth multi-burner ameless oxidation furnace is
situated in the laboratory of Process and Energy. With this test furnace it is possible to study the impact
on the combustion behaviour of many dierent burner arrangements, with dierent distance between
burners or between burner and cooling tubes. In this way unique studies on ameless combustion
in multi-burner conguration are possible. In this contribution, the impact of burner positioning and
operating modes are presented in order to identify an optimal conguration of the three regenerative
burner pairs to maximize the energy eciency and minimize the pollutants emissions. Also, to widen
the range of applicability of ameless combustion technology, the impact of changes in fuel and
315
316
Chapter 17 E.-S. Cho, B. Danon and W. de Jong
Figure 17.1: Schematic diagram of the multi-burner excess enthalpy combustion (MEEC) furnace.
oxidizer composition on the combustion process in a furnace operated using ameless combustion
will be investigated. Natural gas being the base-case fuel, also the combustion of biogas and several
low caloric gas mixtures of relevance for the iron and steel making and the petrochemical industry,
will be investigated. By combining detailed measurements and modeling in multi-burner ameless
oxidation furnaces insight will be gained in the ame structure, heat transfer enhancement and the
emission reduction.
17.2
MEEC furnace
A furnace equipped with regenerative FLOXT M burners has been designed, built and commissioned at
Delft University of Technology. Figure 17.1 shows a schematic diagram of the multi-burner furnace. It
consists of three pairs of regenerative FLOXT M burners, each with a rated thermal power of 100 kWth ,
thus 300 kWth in total. The furnace has inner dimensions of 1500 x 1500 x 1850 mm (length x width
x height). The insulation consists of three layers of ceramic bricks, together 300 mm thick. During
the experiments performed sofar the temperature in the furnace was measured at various locations
with S-type thermocouples. One of those measurements, a double thermocouple in the side wall, was
determined to characterize the furnace temperature. Also, the temperature of the preheated air was
measured in two burners (one burner pair). The fuel and combustion air ow rates are measured by
orice plate dierential pressure meters (Kalinsky Sensor Elektronik, DS2). The combustion air ow
rate is controlled by a manual valve, allowing the variation of the exhaust gas O2 . Eighty percent
of the ue gas is sucked by a fan via the air nozzles over honeycomb ceramics in the burners for
regeneration of the heat, while the remaining twenty percent leaves the furnace directly via the central
stack at the roof. During regeneration the sucked ue gas ows through the ceramic honeycomb heat
exchangers situated inside the burners. The inlet temperature is approximately 950 C and the outlet
temperature is approximately 150 C, under steady state conditions. Thermal load is simulated by a
cooling system which consists of eight single ended concentric tubes, four placed at the bottom of the
furnace and four at the top. Air enters the inner tube, turns at the end and ows back through the outer
tube. This design was made to minimize the temperature gradients along the length of the outer tube,
thus, creating an as uniform as possible heat extraction distribution.
Additionally, an NDIR gas analyzer set (Sick, MAIHAK S710) monitors the ue gas after the
regeneration suction fan (1) and stack (2) for NO and CO concentration in the ue gas. In the same
317
Figure 17.2: Denition of the ring congurations and operating modes.
position the O2 concentration is determined paramagnetically. All the data are stored by a data acquisition system every second. In total 18 anges for the burners are divided over two opposite sides
of the furnace (nine each). In this way, it is possible to investigate dierent burner congurations
in the furnace. Currently, six dierent burner positions (C1-C6) and for each of them two dierent
operating modes (parallel and staggered) are being investigated. Figure 17.2 shows an overview of the
various burner congurations. The large circles represent the burner anges, whereas the small circles
represent the location of the cooling tubes. Three burners are ring, while the other three burners
are regenerating. The lled circles represent the ring burners and the meshed circles represent the
regenerating burners. In the unused burner anges (blank circles) thermocouples are installed.
In parallel mode three burners re at the same side, while in staggered mode two burners re at
one side and one burner at the opposing side. After a certain time interval, all burners switch and
the ring burners start regenerating and vice versa. The burner position can also be divided into two
spatial patterns: triangular (C1, C2, C5) and horizontal (C3, C4, C6). The three pairs of REGEMAT
CD 200 B regenerative FLOXT M burners were manufactured by WS W rmeprozesstechnik GmbH.
a
Each burner has four combustion air/ue gas nozzles (d=20 mm) around a central fuel nozzle (d=12
mm). They can operate in two dierent modes and each pair has a capacity of 90/100 kWth for ame
and ameless operation, respectively.
Figure 17.3 shows a schematic diagram of the burner ring in ame and ameless mode. In ame
mode the air and fuel are mixed in advance and the mixture is injected through the air nozzles only. In
ameless mode the combustion air is injected through the air nozzles and the fuel is injected separately
through the fuel nozzle. During the heating up of the furnace the burners re in ame mode. Once the
furnace temperature exceeds 850 C (this temperature is above the auto-ignition temperature of the
fuel/air mixture) the burners switch to ameless ring mode automatically. Flame mode res with a
318
Figure 17.3: Schematic diagram and ame pictures of burner ring mode for ame and ameless combustion.
Figure 17.4: Control volume and mass and heat uxes of the MEEC furnace.
partially premixed ame and in ameless mode direct fuel injection is applied.
In gures 17.3 (a)-1, 2, 3 the ame shape in ame mode and in 17.3 (b)-1, 2 in ameless mode are
presented. Flame mode shows a blue premixed ame. The burners and insulations turn red during heat
up. At the moment of changing from ame to ameless mode the ames disappear and the furnace
totally turns red of ameless oxidation combustion operation.
Mass and heat balances were calculated over the furnace. Figure 17.4 shows the control volume
of the furnace. For the dierent experiments, the heat and mass balances are calculated under steady
state conditions. Steady state condition is assumed to be reached when the variation of the furnace
temperature is almost zero with respect to time. The steady state temperature of the furnace is between
1080 and 1100 C, depending on the operation parameters.
The mass balance is shown in equation (17.1).
m f uel + mair = mregen + m stack
(17.1)
The ow rates of fuel (m f uel ) and combustion air (owmair ) to the burners are monitored by orice
plates to measure the dierential pressure. The ratio of ue gas that is sucked over the regenerators
(mregen ) and that leaves the furnace via the central stack (m stack ) can be controlled by a manual valve
which is installed before the suction fan. With a vortex ow meter (KOBOLD Messring, PWL)
319
the volume owrate of ue gas sucked over the regenerating burners is measured and subsequently
normalized with a local pressure and temperature measurement. The unknown mass ow rate of the
ue gas through the stack (m stack ) is calculated by conservation of mass. Generally, the regenerated
ue gas (mregen ) ow is about 80 %-wt of the total ue gas ow. The cooling air owrate (mcool ) does
not enter equation (17.1) since there is no mass exchange between the furnace and the heat sink. The
mass owrate of the cooling air is measured for the upper and lower cooling tubes separately with two
thermal ow meters (H ntzsch, TA10). Next, we also consider the heat balance of the furnace. For
o
the calculation of the heat balance equation (17.2) is used.
Q f uel + Qair = Qcool + Qregen + Q stack + Qloss
(17.2)
Heat inputs are the fuels lower heating value (Q f uel ) and the air sensible heat (Qair ). The used
standard fuel is Dutch natural gas. The eciency is dened as the upper and lower cooling tube heat
gain which is calculated by measuring air owrates supplied to the cooling tubes and the dierence
between inlet and outlet temperatures. The upper and lower cooling tubes are measured separately.
Losses are divided into three contributions: the rst is Qregen , the sensible heat in the low temperature
(approx. 150 C) ue gas exiting via the burners, the second is Q stack , the heat in the high temperature
(approx. 1100 C) ue gas exiting via the stack) and the last is Qloss , the heat ux through the furnace
wall and other furnace parts. The last contribution is calculated from the heat balance. The outside
wall temperature varies for dierent operating conditions between 60 C and 80 C under steady state
conditions.
The temperature uniformity in the furnace is an important advantage of ameless combustion.
The spatial uniformity of the temperature in the furnace can be characterized by one variable, the temperature uniformity ratio. It is the normalized root mean square value computed from all temperature
measurement locations (hence perfect uniformity is corresponding to the value zero). It is dened in
equation by [13].
Figure 17.5 shows the temperature measurement positions of the MEEC furnace. Temperatures
are measured in 18 various positions and the thermocouples are protruding around 100 mm from the
inside furnace wall.
Figure 17.5: Temperature measurement positions of the MEEC furnace.
320
Figure 17.6: Evaluation of furnace temperature and emissions during ame to ameless transition.
17.3
17.3.1
Experimental results multi-burner furnace
Flame to ameless transition
This study is focused on the optimization of burner conguration in ameless mode. But it is also
interesting to observe the characteristics of ame to ameless transition during startup of the furnace.
The burner ring modes change automatically from ame to ameless when the furnace temperature is around 850 C. Figure 17.6 shows the characteristics of this transition in the C3 parallel case.
The horizontal axis shows the time where zero (0) indicates the transition point.
During the transition period the furnace temperature slope increases in ameless mode. This
shows the heat transfer enhancement based on the high momentum injection method since the ame
spreads all around the furnace. Additionally, the fuel owrates increase slightly (accompanied by a
decreasing exhaust gas O2 ) in ameless mode which also increases the heat transfer. The most interesting observation is the trend in the emissions. Especially, NO emission sharply drops in ameless
mode from 80 ppm to less than 10 ppm because combustion takes places in a highly diluted mixture
which decreases the ame temperature. NO emissions dominantly depend on the temperature. It
shows that the FLOXT M combustion method is an eective technique to achieve low NO emissions.
Also, CO emissions are decreased in ameless mode. High momentum injection enhances the fuel
and air mixing which decreases the concentration of unburned products such as CO even at relatively
low ame temperature. From now on, the description of the experiments concerns the ameless mode.
The data are collected under steady state conditions and averaged over a period of ten minutes. Whenever changing the various operating conditions such as exhaust gas O2 percentage, operating mode
(parallel, staggered), etc. care has been taken to report data collected after the steady state conditions
have been reestablished.
321
Figure 17.7: Furnace temperature prole for dierent cycle times.
Figure 17.8: Time evolution of regenerator temperature for dierent cycle times.
17.3.2
Temperature variation for cycle time
The furnace temperature variation with cycle time is shown in Figure 17.7. The furnace temperature
slightly increases with increasing cycle time. Also, the uctuations are higher in the temperature for
longer cycle times.
Figure 17.8 shows the temperature characteristics of the regenerator installed inside the burner.
The temperature variation over a cycle time is increasing with longer cycle times, because the total
heat added or lost during a cycle time increases. The maximum regenerator temperature is hardly
increasing but the minimum temperature is decreasing. The temperature dierence is 153 C in 20s
and 214 C in 60s. The regenerator exit temperature is higher with longer cycle time which slightly
increases the regenerative ue gas loss.
17.3.3
Cooling tube eciency
The eciency of the furnace depends slightly on the congurations and operating modes. Figure 17.9
shows the cooling tube eciencies for the various cases.
322
Figure 17.9: Upper and lower cooling tube eciency of congurations and operating modes.
In all experiments the cooling tube eciency decreases with an increase in the excess air ratio.
This decrease in eciency is related to the increase of the combustion air ow rates which leads
to an increase of the ue gas mass ow rate. Thus, the ue gas energy losses are higher and the
furnace eciency is lower. Also, the furnace temperature slightly decreases with the excess air ratio
increasing. In every conguration the eciency of the upper cooling tubes is higher than that of the
lower cooling tubes. This is explained by the fact that the upper cooling tubes receive relatively more
heat, by convective heat transfer, from the 20 % of the ue gases leaving the furnace via the stack,
situated in the roof. Buoyancy eects may also play a role. The dierence is about 1 2 %. However,
in case C2 a larger dierence between the upper and lower cooling tubes is observed. In this case, the
lower three burners are installed at the middle level, instead of at the lower level, as in the other cases,
see also Figure 17.2 This causes the temperature dierence between the upper and lower zone of the
furnace to be somewhat higher in the C2 conguration than in the other congurations, since there is
no burner in the lower furnace zone. The total eciency of case C2 is also lower than that of the other
cases. Comparing between parallel and staggered mode, in all cases the parallel mode shows slightly
higher eciency except for the C2 cases. It shows that in general the parallel ring mode results in
better heat transfer characteristics than the staggered ring mode. The distribution over the dierent
terms in the heat balance equation (2) is only slightly dierent for the dierent burner congurations
and operating modes. Figure 17.10 shows a pie chart of the heat balance contributions in all cases.
The total cooling tube eciency is around 50 %, except for case C2 (45 %). Regenerative ue gas
loss is around 5 % in all cases, stack ue gas loss is 10 to 19% and other losses are 27 to 37%. The
eciencies determined here are lower than values found for a similar furnace, the 200 kWth HiTAC
furnace at KTH, Sweden where around 56 to 65 % was observed [13]. This is probably due to the
fact that in this furnace, the ow rate of cooling air was limited by the capacity of the fan. Also, in
the KTH furnace all of the ue gases are regenerated, thus avoiding the high stack losses due to high
temperature ue gas ows.
17.3.4
Temperature uniformity ratio
Figure 17.11 shows the temperature uniformity ratio for each conguration. Firing in parallel mode
results in a lower temperature uniformity ratio than staggered mode. In parallel mode, three burners
are ring at one side and at the other side three burners are regenerating. This leads to a more stable
ow pattern inside the furnace. On the contrary, in staggered mode, the simultaneous ring burners are
located on both sides. Thus the ow pattern is more complex and the ames also show impinging and
merging phenomena which leads to larger temperature uctuations. Among the parallel ring modes,
323
Figure 17.10: Heat balance of each conguration (C) in parallel (P) and staggered (S) mode.
Figure 17.11: Comparison of the furnace temperature uniformity ratio for each conguration.
congurations C4 and C3 show a lower temperature uniformity ratio than the other congurations.
In C3 and C4 the burners are in one horizontal plane which is better than the triangular arrangement
of C1, C2 and C5. Case C2, with a low eciency, shows a high value of the temperature uniformity
ratio in staggered mode. Thus, temperature uniformity ratio appears also to be of importance for the
furnace eciency.
17.3.5
Emissions
Another important criterion for choosing an optimum conguration is the emission production. Emissions of NO and CO are measured in the various cases under steady state conditions. Actually the
emissions are uctuating in time as a result of the regeneration cycle but here we reported averaged
emissions for a ten minutes period. Figure 17.12 shows the NO and CO emissions for each burner
conguration and operating mode. Each conguration shows a somewhat dierent slope, but similar
trends. NO emissions increase with the excess air ratio, because the O2 percentage increases in the
combustion zone which enhances thermal NO formation. Parallel mode shows lower NO emission
than staggered mode in all congurations. It is also related to the low temperature uniformity ratio
in parallel mode shown in Figure 17.11. There is probably less hot spot formation in parallel ring
mode which causes low production of NO emission. CO emission decreases with an increasing exhaust gas O2 which is an inverse trend compared to NO. Under high excess air ratio conditions, CO is
324
Figure 17.12: NO and CO emissions for congurations and operating modes.
produced in small amounts because the fuel is almost completely combusted. But at lower excess air
ratio conditions the CO emission steeply increases. It may be caused by poor mixing of fuel and air
in the combustion zone and a low residence time in the furnace. The parallel mode generally shows a
lower CO production than the staggered mode. In all cases the parallel mode shows better combustion
characteristics than the staggered mode. This is probably due to the fact that in parallel mode the
temperature distribution is more uniform reducing NO and better mixing is established which reduces
the unburned hydrocarbons.
17.4
Modelling
For an insight in and an explanation of experimental results of furnaces ring HiTAC burners, Computational Fluid Dynamics (CFD) simulations have been widely and succesfully used. The largest
challenges are the modelling of the turbulence-chemistry interaction and nding an optimum between
accuracy and computational times. In most previous CFD simulations of furnaces, combustion has
been modeled using the Eddy Break Up (EBU) model in combination with simple reaction kinetics
[14, 12, 15] or by fast chemistry/assumed PDF calculations [16, 17, 18] with various results.
In this section the simulation of a semi-industrial scale furnace ring two pairs of HiTAC burners
is discussed. Since the Eddy Dissipation Concept (EDC) model seems to be a promising model for
the simulation of HiTAC [19], in the present simulations the EDC model with a skeletal chemistry
mechanism has been applied.
In the following, the process of setting up the simulation, validating it and extracting results from
it are discussed in a chronological order. In general, CFD simulations are based on experimental
results, as is the case here, and thus in section 17.4.1 the experimental setup is presented. Then, the
grid generation is discussed in section 17.4.2 and the applied models in section 17.4.3. After the
(a)
325
(c)
(b)
Figure 17.13: Burner frontview (a), with central rectangular air nozzle and two round fuel nozzles. Furnace
topview (b) with encircled numbers indicating the probe insertion positions. Furnace sideview (c) with indication of probe measurement heights and cooling tube diameter and positions. All measures are in mm.
(a) Parallel
(b) Staggered
Figure 17.14: Firing modes. Parallel (a): A and B ring, while C and D regenerating, et vice versa. Staggered
(b): A and D ring, while B and C regenerating, et vice versa.
validation of the simulations in section 17.4.4, some of the obtained results are presented in the last
section, as an example of how CFD simulation can be utilized.
17.4.1
Experimental furnace
Experiments have been performed on a HiTAC semi-industrial test facility at KTH in Stockholm,
Sweden, see Figure 17.13. The internal dimensions of the furnace are 1600 x 1600 x 2900 mm,
for width, height and length, respectively. The rebrick insulation is 300 mm thick. In the furnace
side walls two pairs of regenerative NFK-HRS-DF burners (manufactured by Nippon Furnace Koygo
Kaisha) are installed. These burners have two fuel nozzles, separated 0.185 m from each other, with a
central rectangular air nozzle in the middle. Each burner pair has a thermal capacity of 100 kWth . The
furnace is cooled with air owing through four horizontal cooling tubes, made of a special temperature
resistant kanthal alloy.
The burners have ceramic honeycomb heat exchangers incorporated. The ue gas is sucked by a
fan via the air nozzles over these honeycombs (thus heating them) for regeneration of the heat. Two
burners are ring simultaneously, while the other two burners are regenerating. After a time interval of
30 seconds they switch and the ring burners start regenerating, et vice versa. Dierent ring modes.
i.e. selection of burners ring simultaneously, can be chosen, see Figure 17.14.
The fuel consisted of 90%-vol methane, 9%-vol higher hydrocarbons up to hexane and 1%-vol
inerts (CO2 and N2 ). The wall temperature measured in the middle of the roof of the furnace was
326
dened as the furnace temperature (Tfurn ). This temperature was kept constant at a preset value by
varying the amount of cooling air owing through the cooling tubes.
The objective of the experiments and simulations was (1) to observe dierences in the performance
of the furnace operating in two dierent ring modes, parallel and staggered, and (2) to explain these
dierences using the detailed CFD simulations.
17.4.2
Mesh
As a rst step, a mesh (or grid) is generated for the furnace. In general (industrial) furnaces have
simple geometries (mostly a rectangular box). This is an advantage for meshing, since the mesh
can be generated with only hexahedral cells. At the other hand, industrial furnaces have both large
measures (total sizes are typically in the order of 2 to 50 meters), in combination with very small parts
(e.g. burner nozzles have a diameter of typically 10 to 40 millimeters). These two orders of magnitude
need to be combined in the mesh gradually, commonly resulting in large meshes, i.e. a high number
of cells.
For this furnace, a three-dimensional mesh was generated using Gambit 2.3. The mesh contained
approximately 750.000 hexahedral cells. Exploiting the symmetry of the furnace with respect to the
midplane through the burners, to reduce the computational time, only half of the furnace was meshed.
In a later stadium, a second mesh with approximately 1.5 million hexahedral cells was generated,
in order to verify the grid-independence of the simulations. Identical boundary conditions and models
were applied using both meshes. Since the results of these two simulations were also identical, it was
concluded that the smaller mesh (of 750.000 cells) could be considered grid-independent.
17.4.3
Models
For all physical phenomena and chemical processes models need to be selected for the simulation.
When selecting these models, a balance needs to be found between accuracy and computational time.
In many cases more accuracy means longer computational time.
In general, RANS (Reynolds Averaged Navier-Stokes) modelling is applied for turbulence closure
for large meshes. The most widely applied turbulence models are the (realizable) k- model and the
Reynolds Stress model. In this particular furnace the realizable k- model was selected, because it is
shown to give improved results, compared to the standard k- model, for the prediction of jets [20], as
is the case in this furnace. Currently, computer power is at the edge of suciency for the application
of Large Eddy Simulation (LES) in large meshes.
For the combustion of the natural gas nite rate chemistry is applied, in particular the Kee mechanism for the combustion of methane [21]. This mechanism comprises 16 dierent species and 46
Arrhenius type reactions. The actual fuel used in the experiments was natural gas, consisting mainly
of methane, but also around 9%-vol of higher hydrocarbons. Therefore, the natural gas was simulated with a simulation gas, containing only methane. The higher hydrocarbons present in the natural
gas were replaced by methane, in such way that the heat release per unit mass remained equal. The
turbulence-chemistry interaction is taken into account by using the Eddy Dissipation Concept model
[22]. In this model, the extend of the nite rate reactions is coupled to the turbulence intensity.
Radiative heat transfer is accounted for using the Discrete Ordinates method and the cell-based
Weighted Sum of Gray Gases model for the absorption coecient of the gas mixture. The emissivity
of steel was set to 0.8 and that of the rebrick insulation to 0.68.
Also, for the physical properties of the media many dierent models are available, and again
both the accuracy and computational time needs to be considered. In this simulation, the density of
327
the mixture is calculated by the multi-component ideal gas law. The molecular viscosity and thermal
conductivity are taken constant. The mass diusivity is calculated with the constant dilute approximation. Finally, the heat capacities are determined per species with a temperature dependent piece-wise
polynomial, and for the mixture with the mixing law.
Next, the boundary conditions need to be dened. Mass ow rates, gas compositions, pressures
and temperatures of the incoming uids are set to values derived from the experiments. For the outows of the furnace pressure outlets were used, which means that the pressure is set to be 1 atmosphere
at the outlet.
Furthermore, the heat losses through the walls need to be incorporated in these type of simulations.
Simple methods are setting a constant wall temperature (e.g. a measured wall temperature) or a
constant heat ux through the wall. The drawback of these methods is that the wall will be taken
uniform (a uniform temperature or heat ux), which in reality is not the case. In reality hotter and
cooler zones will be present on the wall. Also, these dierent zones will also have dierent heat
uxes to the environment. Therefore, it was decided to incorporate the heat loss through the wall
by calculating the convective and radiative heat ux to the environment in these simulations. For a
realistic value of the eective thermal conductivity of the walls, the predicted furnace temperature was
tuned on the measured furnace temperature.
Another important factor for the heat transfer from a uid to a solid is the near-wall modelling.
Fluent has the capability of enhanced wall treatment, which is a two-layer model combined with
enhanced wall functions. Fluent checks the wall y+ values and chooses on its value which type of
near-wall modelling it uses.
The NO formation is generally studied in post-processing mode. This means all other transport
equations (mass, heat, radiation etc.) are not calculated, thus their values stay constant. In this furnace,
calculations have been performed both simultaneously and separately for thermal NO, prompt NO,
N2 O intermediate and NO reburning mechanisms. The standard Fluent models were applied. The
uctuations in the mean temperature and O2 concentration have been included in the mean reaction
rate of NO by a two-variable PDF which is assumed to be a two-moment beta function. The inuence
of the uctuations on the total NO formation rate have been investigated and turned out to be small.
Calculating the small mesh, containing 750.000 cells, and with the above mentioned models and
boundary settings, the computational time for 1000 iterations was about 3 hours, running parallel on
a 8-core cluster. Full convergence was reached after about 50.000 iterations.
17.4.4
Validation
Any simulation needs to be validated. For furnace simulation in general in-furnace measurements
are used for validation. These in-furnace measurements can be temperatures, species concentrations
(e.g. O2 , CO2 , NO, CO, UHC (Unburned HydroCarbons)), uid velocities, etc. It is important for
good validation that the (experimental) data used for the validation is not priorly used as a boundary
condition for the simulation.
In this case, the simulation was validated with both in-furnace measurements (temperature and
O2 ) and with the overall heat balance.In Figure 17.15 the predicted values for temperature and O2
concentration inside the furnace are compared with the measured values. The hole number is denoted
on the x-axis, with four points per hole, representing the four dierent heights, from low to high, where
measurements were performed, see also Figure 17.13(b) and (c). No measurements were performed
in hole 1.
The predicted temperatures inside the furnace agree well with the measured values. Also, the
trend in the temperature agrees reasonably.
328

Parallel
Staggered
10
CFD
10
CFD
Position ()
H13
400
H12
H11
H9
600
H10
H8
800
H7
H6
1000
Temperature (C)
Oxygen (%vol dry)
1200
H3
H13
H12
400
Experiments
H11
600
H10
H9
800
H8
H7
1000
H6
H3
1200
Temperature (C)
1400
Experiments
Oxygen (%vol dry)
1400
Position ()
Figure 17.15: Comparison between the steady state experimental and simulated results for parallel and staggered mode. Squares represent the experiments, solid lines the CFD simulations. Oxygen concentrations (%-vol
dry) are in gray and temperatures (C) are in black.
Table 17.1: Heat balance comparison between experiments and simulations.
Cooling tubes (kW)

Flue gas losses (kW)
Wall losses (kW)
Total (kW)
Parallel
Experiment CFD
91.2
67.4
29.9
50.0
70.0
77.3
191.1
194.7
Staggered
Experiment CFD
76.4
61.9
30.7
48.5
83.9
83.1
191.0
193.5
For the staggered case, the O2 concentrations agree well with the measurements. However, in
parallel mode, the predicted O2 concentrations are consistently lower than the measured values. This is
due to air leakage. During the experiments in parallel mode, the furnace pressure dropped signicantly
below zero for two seconds every cycle. Since it is known that there are openings and cracks in the
furnace, during these periods of negative pressure, ambient air leaks into the furnace. This air leakage
was proven by simple calculations to be large enough to explain the observed dierences. In staggered
ring mode the negative pressure peaks were negligible.
In Table 17.1 the heat balance of the experiments and simulations are compared. The heat extracted by the cooling tubes in the simulations is lower than the experimental values. This non extracted heat is leaving the furnace partly via the walls and partly as sensible heat of the ue gas. This
remains an unresolved issue in the simulations. However, it is noted that the trend in the results for
parallel and staggered ring mode do agree, showing that essential aspects of the change of ring
mode are captured.
It was concluded that the CFD simulations are validated suciently for the purpose of investigating the observed dierences in eciency and emissions for the dierent ring modes.
17.4.5
Results
With the two validated simulations of the furnace, ring in both ring modes, it was possible to investigate the observed dierences in heat extraction of the cooling tubes and production of NO. In this
Radiative heat extraction (W/m2)

4
x 10
329
Cooling tube 1
Cooling tube 2
x 10
Parallel
Staggered
5
0
1
2
zdistance (m)
5
0
Parallel
Staggered
1
2
zdistance (m)
Figure 17.16: Mean radiative heat extraction (W/m2 ) along the cooling tubes.
Figure 17.17: Higher values of the normalized emissive source term (-) on the furnace midplane. The horizontal
gray lines denote the centrelines of the cooling tubes, cooling tube 1 is the lower and cooling tube 2 is the upper
line.
section the highlights of the numerical results are presented, as an example of how CFD simulations
can be utilized for further insights in HiTAC furnace behaviour. For further details see [23].
Heat extraction
The dierence in the heat extraction of the cooling tubes turned out to be mainly due to an increased
radiative heat extraction in parallel mode. Therefore, the mean radiative heat extraction (RHE) by
the cooling tubes, plotted along the tubes, is presented in Figure 17.16. Additionally, the normalized
emissive source term (Re ), dened in Equation 17.3, was considered [24],
Re =
T
r T r
(17.3)
where is the absorption coecient (m1 ), T the temperature (K) and the subscript r denotes reference
values (r = 1m1 , T r = 298K). In Figure 17.17 contours of higher values of the normalized emissive
source term (Re
750) in the furnace midplane for both ring modes are presented. These zones
indicate where most radiative energy is emitted by the hot gases. The centerline of cooling tube 1 is
the lower gray line in Figure 17.17 and cooling tube 2 the upper gray line. The z-distance increases
from left to right in Figure 17.17.
In Figure 17.16 it is observed that cooling tube 1 has comparable radiative heat extraction (RHE)
for both ring modes. It is noted that an increase in the RHE is observed in staggered mode around
330
Table 17.2: NOx concentrations in the experiments and the simulations.
Experiment
Simulation
Unit
ppmv @3%O2 dry
ppmv @3%O2 dry
Parallel
39
48
Staggered
53
86
z-distance 1.5 m, which is due to the location of high Re in this location, see also Figure 17.17. Note
that the central air nozzles of burners B and D are at z-distance 1.45 m, see also Figure 17.13(b).
For cooling tube 2 the RHE is higher for parallel mode along the entire length of the tube. This
is conrmed in Figure 17.17 by the location of the higher values of Re in parallel mode. There is no
distinct increase visible in the RHE at z-distance 0.5 m (centerline of burners A and C) for staggered
mode in cooling tube 2. This is due to the fact that the cooling air is entering at z-distance 2.9 m and
then ows towards 0 m. Therefore, the RHE decreases with decreasing z-distance due to the increasing
temperature of the cooling air. The expected increase in RHE of cooling tube 2 in staggered mode at
z-distance 0.5 m cancels with the decrease due to this phenomenon.
Overall, it is observed in Figure 17.16 that in parallel mode more radiative heat is extracted by
the cooling tubes. This is conrmed in Figure 17.17 by an overall larger zone with high values of the
emissive source term (Re ) in parallel mode. In staggered mode higher maxima in the Re are observed
in the center of the furnace, however, these maxima are located further from the cooling tubes.
NO production
In Table 17.2 the measured and simulated values for the concentration of NOx in the ue gas are compared. The simulated values for NOx are slightly overpredicted. However, their order of magnitude
do agree.
From detailed analysis of the NOx predictions in combination with the ow patterns in the furnace,
it turned out that the dierence in production of NO in the two ring modes was linked to the merging
behaviour of the comburant jets, see also [25]. In order to illustrate this correlation between the jet
merging behaviour and the NO production, the maximum reaction rate of NO at x/d = 12 is plotted
versus the location of the end of the merging zone (Lm ) divided by the air nozzle diameter (d) for all
fuel nozzles in Figure 17.18. It is observed that the slower the fuel jet merges with the air jet (higher
values of Lm /d), the less NO is formed. The correlation seems to be linear, although more data would
be required to verify this.
Another interesting nding in the NOx predictions was that there was no production of NO via
the prompt NO mechanism, while there was NO depletion via the NO reburning pathway. This is
noteworthy, since both the prompt NO and NO reburning mechanisms occur in fuel-rich zones, where
CHi radicals react with N2 and NO, respectively. Since the fuel and air are introduced separately,
a fuel-rich zone is present in front of the fuel nozzles. The reason that in these zones no NO is
formed via the prompt mechanism but NO is reburning, is the temperature dependence of the chemical
reactions. In Figure 17.19 the reaction rates for prompt NO and NO reburning are plotted versus the
local temperature. Indeed, it is noted that at lower temperatures the NO reburning mechanism shows
higher reaction rates than the prompt NO mechanism.
17.4.6
Conclusions
Industrial furnaces operating in HiTAC (or FLOX, MILD, etc.) are very suitable for detailed CFD
simulation. The largest challenge is that they generally need large meshes and thus have long com-
331
Maximum NO rate (kmol/m3s)
x 10
6
4
2
Parallel
Staggered
0
8
10
11
Lm/d ()
12
13
Figure 17.18: Maximum reaction rate of NO at x/d = 12 versus the location of the end of the merging zone
Lm /d (-) for all fuel jets.
0
Reaction rate (mol/m3s)
10
10
10
20
10
30
10
Prompt NO
NO reburning
40
10
500
1000
Temperature (K)
1500
2000
Figure 17.19: Rate of formation of prompt NO and rate of destruction for NO reburning (mol/m3 s) versus the
local temperature (K). Data are values of all grid cells of the furnace midplane.
putational times. Therefore, most furnace simulations are currently performed with relatively simple
models for turbulence, chemistry and their interaction. The progress in computer science will allow
more advanced models to be used in the future.
However, with the models presently available, it was shown that useful results can be obtained.
Also the HiTAC type of concept could be reproduced with these models. Especially, for the comparison of dierent operating conditions and/or parametric studies of furnaces with HiTAC burners, CFD
simulations can be a helpful tool.
17.4.7
Acknowledgement
This work performed in the framework of the project 06910 Dynamic behaviour of a Multi-burner
Excess Enthalpy Combustion (MEEC) system for industrial process furnaces is nancially supported
by the Technology Foundation STW and the Dutch Flame Foundation NVV of the Netherlands.
References
[1] C.E. Baukal, R.E. Schwartz, The John Zink Combustion Handbook, CRC Press (2001).
332
[2] J.A. Wuenning, J.G. Wuenning, Progress in Energy and Combustion Science 23 (1997) 8194.
[3] M. Katsuki, T. Hasegawa Proceedings of the Combustion Institute 27 (1998) 31353146.
[4] H. Tsuji, A.K. Gupta, T.Hasegawa, M. Katsuki, K. Kishimoto, M. Morita, High Temperature Air
Combustion, CRC Press (2003).
[5] I.B. Ozdemir, N.Peters, Experiments in Fluids 30 (2001) 683695.
[6] A. Cavaliere, M. de Joannon, Progress in Energy and Combustion Science 30 (2004) 329366.
[7] E.-S. Cho, S.H. Chung, Journal of Mechanical Science & Technology 23 (2009) 650658.
[8] A. Milani, A. Saponaro, IFRF Combustion Journal(2001) Article Number 200101 132.
[9] J.A. Miller, C.T. Bowman, Progress in Energy and Combustion Science 15 (1989) 287338.
[10] T. Plessing, N. Peters, J.G. Wuenning, Proceedings of the Combustion Institute 27 (1998) 3197
3204.
[11] H. Kobayashi, K. Oono, E.-S. Cho, H. Hagiwara, Y. Ogami, T. Niioka, JSME Int.J. 48 (2006)
286292.
[12] W. Yang, M. Mortberg, W. Blasiak, Scandinavian Journal of Metallurgy 34 (2005) 715.
[13] N. Radi, W. Blasiak, Applied Thermal Engineering 26 (2006) 20272034.
[14] W. Yang, W. Blasiak, IFRF Combustion Journal(2006) Article Number 200603.
[15] R.H. Hekkens, M. Mancini, Non-isothermal CFD model fo the HEC burner and furnace. Technical Report G108/y/2, IFRF Research Station (2004).
[16] T. Ishii, C. Zhang, S. Sugiyama, Journal of Energy Resources Technology 120 (1998) 276284.
[17] T. Ishii, C. Zhang, Y. Hino, Heat Transfer Engineering 23 (2002) 2333.
[18] R.H. Hekkens, Non-isothermal CFD model of the HEC burner and furnace (additional calculation). Technical Report G108/y/3, IFRF Research Station (2004).
[19] B. Danon, B. Venneker, W. de Jong, D.J.E.M. Roekaerts, VDI Berichte 1988 (2007) 339344.
[20] T.-H. Shih, W. Liou, A. Shabbir, Z. Yang, J. Zhu, Computers Fluids 24 (1995) 227238.
[21] R. J. Kee, F. M. Rupley, J. A. Miller, M. E. Coltrin, J. F. Grcar, E. Meeks, H. K. Moat, A. E.
Lutz, G. Dixon-Lewis, M.D. Smooke, J. Warnatz, G. H. Evans, R. S. Larson, R. E. Mitchell, L.
R. Petzold, W. C. Reynolds, M. Caracotsios, W. E. Stewart, P. Glarborg, C.Wang, O. Adigun, W.
G. Houf, C. P. Chou, S. F. Miller, P. Ho,D. J. Young, Chemkin v. 4.0. Technical report, Reaction
Design Inc. (2004).
[22] I.S. Ertesvg, B.F.Magnussen, Combustion Science and Technology 159 (2000) 213235.
[23] B. Danon, A. Swiderski, W. de Jong, W. Yang, D.J.E.M. Roekaerts, submitted to Combustion
Science and Technology (2010).
[24] M.F. Modest, Radiative Heat Transfer. Academic press, 2nd edition (2003).
[25] A. Vouros, Th. Panidis, Experimental Thermal and Fluid Science 32 (2008) 14551467.
Chapter 18
New combustion concepts in Diesel engines

L.M.T. Somers
18.1
Introduction to Diesel engines
Current legislation (EURO 6, US 2010) limits emissions to extremely low NOx (US2010: 0.27 g/kWh)
and soot (0.013 g/kWh) levels. Contemporary diesel combustion technology results in engine-out
emissions that are much higher than these limits. Therefore, aftertreatment systems like SCR (Selective Catalytic Reduction) and DPF (Diesel Particulate Filter) technologies have to be used for the
reduction of NOx and soot. As such any development in combustion concepts that lowers NOx /soot
emissions will reduce the demands for the aftertreatment system required and thus minimizes related
costs. To meet these demands much research is initiated that rely on the conception that engine-out
emissions are closely related to the realized (-T )-state in the engine, like illustrated in gure 18.1.
Therefore to overcome the limits, approaches that lower the temperature and enhance mixing are
explored. New combustion concepts like HCCI (homogeneous charge compression ignition) [11][13], PCCI (partially premixed charge compression ignition) [14, 15] and MK (modied kinetics)
Figure 18.1: Distinction between engine test points that exhibited Low-Temperature Smokeless Combustion
and those that showed High Smoke [1].
333
334
[16, 17], try to lower the maximal local temperature and equivalence ratio by allowing a (large) premixed phase by either very early (HCCI/PCCI) or/and late (MK) injection strategies. In the upcoming
sections more detail is given on all of these concepts.
18.2
Combustion modes in modern Diesel engines
To put the new combustion concepts into perspective the classical Diesel combustion concept is explained rst. Internal combustion engines basically come in two avors. One is the Spark Ignitionengine (SI-engine) or Otto-engine and the other one the Compression Ignition-engine (CI-engine) or
Diesel-engine. As their names suggest the principal dierence is the way in which the fuel is ignited. SI-engines are operated such that a more or less pre-mixed mixture is provided to the engine
and combustion is initiated and controlled by the spark timing. After initiation a turbulent premixed
ame travels through the combustion chamber. CI-engines rely on auto-ignition and combustion is
controlled by the timing and duration of the fuel delivery (injection-timing). After auto-ignition a turbulent (lifted) diusion ame develops that remains up to the completion of the injection event. Diesel
engines have been around since the late 19th century when Rudolf Diesel (1858-1913) developed his
rst prototype in 1893 and developed it till 18971 [28]. A major step again was made when Robert
Bosch2 (1861-1942) invented the rst high pressure fuel injection (inline-pump) system in 1927 [27],
see g 18.15. Using these high-pressure pumps the problem of adequate fuel delivery and atomization
within the short period available in each cycle was solved. It remained the de-facto standard in Diesel
engines up to the late 90s of the previous century as electronically controlled common-rail systems3
emerged on the market and are now more or less the standard. This development, however, opened up
a new area of research due to the exibility of the CR-systems.
18.2.1
Spray combustion
In modern Diesel engines fuel is injected directly into the cylinder using high-pressures such that the
fuel vaporizes quickly and mixes with the air. Before the injection stops the mixture will auto-ignite
and an unsteady lifted diusion ame establishes (the mature stage). The so-called mature diesel
spray ame structure according to Decs conceptual model [26, 29] is depicted in gure 18.2. The
fuel is injected and it atomizes, vaporizes and mixes with the surrounding hot air. At the leading
edge of the lifted spray ame a rich ( = [2 6]) premixed ame establishes which is prone to
form Soot (Initial Soot Formation). Soot formation continues in the hot core region which is depleted
of oxygen. The partially oxidized fuel/CO mixture continues to move to the outer edges where a
diusion ame completes the fuel conversion. It is in this zone where NOx is formed due to the high
local temperatures.
18.2.2
Low temperature combustion
LTC [1]-[8], i.e. Low-Temperature-Combustion, tries to lower the ame temperature in the engines
T -map. For a given engine load-point4 this can be achieved by creating a higher dilution using
higher inlet pressure (more air) and/or apply exhaust gas recirculation (EGR). The latter also increases
1
Given the opportunity by Heinrich von Buz, director of MAN AG

The founder of Robert Bosch GmbH, the worlds largest supplier of automobile components > 40Geuro yearly sales
3
Alfa Romeos JTD engines
4
This is essentially dened by the amount of fuel delivered to the cylinder, given the thermal eciency isnt changed too
much
2
335
Thermal NO Production Zone
Liquid Fuel
Diffusion Flame
Rich Fuel/Air Mixture

Fuel-Rich Premixed Combustion
Soot Oxidation Zone
Initial Soot Formation
10
Scale (mm)
20
Low
High
Soot Concentration
Figure 18.2: J.E. Decs conceptual model of Diesel Combustion: the mature stage
the specic heat-capacity of the mixture which has been used for some time already to reduce NOx
normally at the cost of an increase in Soot levels. LTC aka smokeless diesel combustion, however,
pushes this to a new level. By using ultra-high EGR levels the ame temperature is reduced such that
time and temperature are not sucient to form soot altogether [1]. This concept is nicely illustrated
by the luminosity pictures shown in (see gure 18.3). Initially the smoke increases with increasing
EGR level, but at a certain point the smoke begins to decrease again.
Figure 18.3: A Method for Low Temperature Combustion using high EGR rates to reduce ame temperatures
suciently that smoke is reduced. Images show ame luminosity from soot [1]. Note that Air-Fuel ratio (AFR)
is linked to the equivalence ratio = AFR stoi /AFR, AFR stoi 14.5.
336
18.2.3

HCCI/PCCI
Both acronyms HCCI (Homogeneous Charge Compression Ignition) and PCCI (Premixed Charge
Compression Ignition) are used in literature to indicate strategies that primarily try to enhance mixing
of fuel prior to ignition. Even more so, normally the qualication H/PCCI is reserved for approaches
where injection has nished (well) before the auto-ignition event occurs. Within the T -map the
major strive is to reach the region left of the Soot-formation region for PCCI. HCCI is a limiting
situation where injection is so early that the mixture is nearly homogeneous. As Diesel engines are
normally operated lean the region for HCCI will extend below the = 1 line. However, the current
conception is that pure HCCI will not be a viable approach for the complete operating range in the
Diesel engine. This is mainly due to practical limitations regarding the maximal pressure rise rate limit
and CO/UHC5 emissions. Therefore PCCI is currently studied by many researchers. The working
hypothesis in this eld is to create a decoupling of the ignition from the injection process suciently
such that rich zones are avoided yet stay away from the conditions where excessive NOx is formed.
18.3
PCCI/LTC studies at the TU/e: Experimental research
In this section a short overview is presented on experimental studies that are performed at the TU/e.
For more detail and more results the reader is kindly referred to the theses [45, 43]. All of the results
in this part are from the latter two studies.
To enhance the readability some enginespeak is unavoidable and although not complete the most
common ones are introduced here:
SOI Start of Injection: Start of the actual fuel delivery
EOI End of Injection
SOA Start of Actuation: Start of actuation of the injector. This is not equal to the SOI. In our engine
a typical dierence at 1200 RPM is SOI-SOA 4 CA .
CAx with x a number between 0-100. In this research CAx refers to the crankangle at which x% of
the total heat is released. In practice this is obtained from a so-called heat-release analysis using
the in-cylinder pressure. Common ones are CA5,CA10 to indicate start of ignition, CA50 to
classify the combustion phasing.
Dwell CA5-EOI, CA10-EOI, if positive injection has ended before combustion starts.
CD,EID CA50-SOI, the Combustion Delay or Engine Ignition Delay.
RPM,REVS RPM, Revolutions Per Minute, REVS=RPM/60
CR Compression Ratio, the ratio of largest and smallest volume during the engine cycle CR =
with V s the stroke volume and Vc the so-called clearance volume.
V s +Vc
Vc
IMEP Indicated Mean Eective pressure. Computed from either the torque that is measured IMEP=2/V s
or computed from the in-cylinder pressure trace IMEP= pdV/V s . It is a measure to qualify the
engines load-point irrespective of the stroke volume.
SE Specic Emission is dened as g of emissions per energy output Work commonly measured in
kWhr!. S Ei = mi /(IMEP V s REVS) 1000/3600 [g/kWhr] if [mi ] = [g/s]
Unburned Hydro-Carbons

18.3.1
337
Experimental apparatus: the CYCLOPS
The CYCLOPS6 is a dedicated engine test rig, Table 18.1, [40] and [41], designed and built at the
TU/e, based on a DAF XE 355 C engine. Cylinders 4 through 6 of this inline 6 cylinder HDDI
engine operate under the stock DAF engine control unit and together with a water-cooled, eddycurrent Schenck W450 dynamometer they are only used to control the crankshaft rotational speed of
the test cylinder, i.e. cylinder 1.
Table 18.1: Cyclops Specications
Base Engine
Cylinder
Bore (mm)
Stroke (mm)
Compression Ratio
Bowl Shape
Bowl Diameter
6 cylinder HDDI diesel

1 cylinder for combustion
130
158
12:1
M-shaped
100
During engine warm up or in between measurements, the CYCLOPS is only red on the three
propelling cylinders. Once warmed up and operating at the desired engine speed, combustion phenomena and emission formation can be studied in the test cylinder. Apart for the mutual cam- and
crankshaft and the lubrication and coolant circuits, the test cylinder operates autonomously.
Fed by an Atlas-Copco air compressor, the intake air pressure of the test cylinder can be boosted
up to 5 bar. The fresh air mass ow is measured with a Micro Motion Coriolis mass ow meter. Nonring cylinders 2 and 3 act as EGR pump cylinders, gure 18.1, the purpose of which is to generate
adequate EGR ow, even at 5 bar charge pressure and recirculation levels in excess of 70%. The EGR
ow can be cooled both up- and downstream of the pump cylinders. Several surge tanks, to dampen
oscillations and ensure adequate mixing of fresh air and EGR ows, and pressure relief valves, to
guard for excessive pressure in the circuit, have been included in the design.
Fuel to cylinder 1 is provided by a double-acting air-driven Resato HPU200-625-2 pump, which
can deliver a fuel pressure up to 4200 bar. To dampen pressure uctuations originating from the pump,
two accumulators are mounted downstream of the pump of which the second, placed near ( 0.2 m)
the fuel injector is to mimic the volume of a typical common rail. The prototype common rail injector
used can inject the fuel with a pressure up to 3000 bar and has a nozzle with 8 holes of 0.151 mm
diameter with a cone angle of 153 degrees. The steady state fuel mass ow is measured with a Micro
Motion mass ow meter, while the time resolved pressure and temperature are measured in the fuel
line approximately 50 mm from the injector with a Kistler pressure sensor. Furthermore the injector
actuation current is measured with a clamp meter.
For measuring gaseous exhaust emissions a Horiba Mexa 7100 DEGR emission measurement
system is used and exhaust smoke level (in Filter Smoke Number or FSN units) is measured by
an AVL 415 smoke-meter. Furthermore, the engine is equipped with all common engine sensors,
such as intake and exhaust pressures and temperatures, and oil and water temperature. All of these
quasi steady-state engine data, together with air and fuel ows are recorded by means of an in house
data acquisition system (TUeDACS), see [42]. Finally, a SMETEC Combi crank angle resolved data
6
This is not an acronym, it just refers to the one-eyed giant from Greek mythology. In its original design the engine was
prepared for optical access through a single hole using an endoscope.
338
Figure 18.4: Test section of the Cyclops engine test rig. Cylinders 4-6 propel the testcylinder T and the (EGR)
pump cylinders 2-3.
acquisition system, see [40] and [41], is used to record and process cylinder pressure (measured with
an AVL GU12C uncooled pressure transducer), intake pressure, fuel pressure and temperature and
injector current.
18.3.2
PCCI experiments
Engine measures
In a short term scenario, PCCI combustion will be used in the low load part of the engine operating range only, with conventional CI combustion at higher loads. This scenario implies that engine
hardware design would be very close to that of current modern diesel engines, with compression ratio
possibly made load dependent through implementation of variable valve actuation (VVA). Therefore
optimizing operating conditions should focus on parameters like EGR level, intake temperature and
intake pressure. All of these measures aect the temperature of the cylinder charge (lowering it) such
that for a given fuel ignition is delayed.
Compression Ratio determines the temperature reached during compression of the cylinder charge7 .
Clearly (see g. 18.5) for the higher compression ratio PCCI combustion regime cannot be
reached without EGR. The so-called Dwell, dierence between End of Injection (EOI) and
10% heatrelease (CA10) is always negative meaning combustion has started before injection
has ended. As such ultra low NOx values will not be reached as shown in the right plot in gure 18.5. The fact that the specic emissions rise with decreasing CA50 is due to the decreasing
thermal eciency if the timing is advanced.
7
Remember adiabatic compression pV = c or equivalently T V 1 = c
339
Diesel, Pin 1.25, no EGR, 850 s actuaon
40
Diesel, Pin 1.25, no EGR, 850 s actuaon
50
=15:1
=12:1
40
=12:1
ISNO [g/k
Ox kWh]
Ignion Dwell [CA]
30
20
10
0
-10
=15:1
30
20
10
0
-10
0
10
CA50 [CA aTDC]
20
-10
0
10
CA50 [CA aTDC]
20
Figure 18.5: Left: Dwell (CA10-EOI) as a function of combustion phasing (CA50) for dierent compression
ratios.
EGR level has a signicant eect on the temperature in the cylinder due to the eect of the higher
heat capacity of exhaust gases (H2 O) mainly (see appendix for an illustration, gure 18.16).
For NOx there seems to be a very strong correlation between the combustion delay (EID/CD)
and the emission level as shown in the rightmost plot in gure 18.6. How this engine delay is
obtained does not seem to matter. This suggests that the level of pre-mixing is the key parameter.
Fuel measures
Another important measure to reach PCCI combustion is improving the ignition quality of the fuel.
The more resistant to auto-ignition (i.e. low cetane number) the more time is available for mixing and
the more the conditions shift to a favorable region in the T -diagram (18.1). Within this perspective the application of certain bio-fuels, like alcohols and second generation bio-mass derived cyclic
oxygenates is intriguing. Results from earlier research with cyclic oxygenates show very promising
results even in the classical combustion regime [20, 21]. Using blends of diesel with these bio-derived
oxygenates [25, 24], might be an extremely interesting concept in combination with high EGR as they
provide a unique platform to reduce CO2 and Soot and NOx at the same time. But even with fossil
CA50 [CA aTDC]
15
10
5
0
15
Diesel, CR=12, Pin 1.05, 850 s actuaon
25
N EGR
No
15wt% EGR
40wt% EGR
60wt% EGR
20
ISNO [g/k
Ox kWh]
20
25
60wt% EGR
40wt% EGR
15wt% EGR
No EGR
No EGR
15wt% EGR
40wt% EGR
60wt% EGR
20
ISNO [g/k
Ox kWh]
25
10
5
15
10
5
-5
-10
-60
-50
-40
-30
-20
SOA [CA aTDC]
-10
-60
-50
-40
-30
-20
SOA [CA aTDC]
-10
10
20
30
40
CD [CA]
50
60
Figure 18.6: For dierent EGR levels (CR=12) Left: Combustion phasing (CA50) as function of SOA. Middle:
specic NOx emissions as as function of SOA. Right: NOx emissions as as function of Engine Ignition Delay
(EID=CA50-SOI).
340
fuels ignition delay can be extended quite simply. Gasoline for instance is especially designed to
prevent auto-ignition as much as possible and has to meet tight quality and renement demands and
consequently its production is costly. Therefore if on a longer term a scenario a markedly dierent
automotive fuel mix would evolve, i.e. low-quality gasolines for PCCI engines, it will have major
impact on the use of the fossil fuel sources and the layout of the reneries. Hence the interest of fuel
companies for PCCI/HCCI Diesel combustion.
Ignition resistivity is enhanced by blending regular Diesel fuel with an aromatic compound (Shellsol
A100). So instead of lowering the charge temperature to delay ignition the fuels resistivity to
ignition is altered. As can be seen in the leftmost picture of gure 18.7 CA50 follows a linear
trend for the late injection events (SOA=[-20 0]) marking the classical combustion regime. For
SOA < 20 this trend is lost and the PCCI-regime is entered. In that region the ignition delay
is closely correlated to the cetane number: for a given SOA it increases if the cetane number
decreases. In this same regime the NOx decreases drastically (middle gure 18.7). If plotted
against the combustion delay (CD) again the it is found that the level of mixing correlates
strongly with the emissions irrespective of how it is achieved.
5
0
-5
30
20
10
0
-60
-50
-40
-30
-20
SOA [CA aTDC]
-10
Diesel
CN=40
CN=30
CN=25
40
-10
-15
CR=15, Pin 2.0, no EGR, 850 s actuaon
50
Di l
Diesel
CN=40
CN=30
CN=25
40
ISNO [g/k
Ox kWh]
CA50 [CA aTDC]
10
50
Diesel
CN=40
CN=30
CN=25
15
ISNO [g/k
Ox kWh]
20
30
20
10
0
-60
-50
-40
-30
-20
SOA [CA aTDC]
-10
10
20
CD [CA]
30
40
Figure 18.7: For dierent fuel blends (CR=12) Left: Combustion phasing (CA50) as function of SOA. Middle:
specic NOx emissions as as function of SOA. Right: NOx emissions as as function of Engine Ignition Delay
(EID=CA50-SOI).
18.3.3
Outlook
Experimental research at the TU/e in this area will continue. On the hardware level new injectors
allowing higher injection pressures and more exible injection strategies are currently installed. In
the near future the engine will also be equipped with a variable valve actuation (VVA) system to be
able to vary the eective compression ratio more easily.
Eects of fuel composition and quality will be another focus point of the research both in PCCI as
well as the classical Diesel combustion regime. There will be a special interest in the applicability of
> 2nd generation bio-fuels as they provide a way to reduce the carbon footprint of the transportation
sector drastically. Note that the Diesel engine will remain the main propulsion for the HD trucks not
withstanding the progress in hybrid and electric powertrain technology.
18.4
PCCI/LTC studies at the TU/e: Numerical research
To study PCCI numerically one could use CFD but then obviously the use of highly detailed reaction mechanism is not straightforward due to the turbulent closures needed. At the TU/e work in
341
this context focusses on the application of tabulation methods like ILDM and FGM. For PCCI-like
combustion a further extension of the method is needed and is currently being investigated. Another
approach is what is commonly referred to as multi-zone modeling. In this approach ow phenomena
are modeled rather rudimentary (turbulent mixing at the most) but (large) detailed mechanisms can
be applied directly. As HCCI/PCCI are mainly governed by auto-ignition chemistry and not ame
propagation it is believed that multi-zone approaches are a viable approximation for this combustion
regime. In the following sections the numerical formulation for this multi-zone approach is explained
and some typical results are presented.
18.4.1
Modeling Approach
The XCCI code, where X represents P/H, is a multi-zone code developed at the TU/e to perform
PCCI /HCCI combustion simulations using a wide variety of detailed chemical models. The multi
zone feature is established by dividing the cylinder volume into a number of zones. In this study the
zones are set in a cylindrical manner and only the outer zone has contact with the cylinder walls. All
of the zones are in contact with the cylinder head and the piston depending on their volumetric ratios.
It must be noted that the geometrical arrangement has most impact on the eect of the cooling model
that is applied. The mathematical and chemical models as well as the initialization of the model are
presented in the following sections of this chapter.
Mathematical model
Energy conservation in the XCCI code is based on the 1st law of thermodynamics for open systems:
dT z
mz cv
=
dt
Ns
i=1
Ns
dmz,i
dVz
+ Qz p
+
ei
dt
dt
Ns
min hi (T in )
z,i
i=1
mout hi (T z ) + DT
z,i
z
(18.1)
i=1
where z is the zone number with Vz its volume. Species are denoted by their index i, with N s in total
and are chracterized by their mass mi,z in each zone. mz = i mi,z is the total mass in zone z, T z the
zone temperature and p is the pressure. The symbols ei and hi ( [Jkg1 ]) denote the specic internal
energy and enthalpy of species i, respectively. Qz is a lumped quantity describing the heat transfer rate
due to heat loss and evaporative cooling due to injection of liquid fuel. min/out is the mass ow through
z,i
the boundaries, with T in the temperature of the incoming ow. The heat capacity of the mixture is
given by the common relation cv = Yi cv,i . Finally the term DT represents a basic turbulent mixing
z
model between the zones similar to that applied by Aceves [30].
The change of mass of the species in each zone is determined by the chemical reactions and the
mass exchange through the boundaries,
dmz,i
= Vz (Mi
dt
Nr
vi j j ) + min/out + Di
z,i
z
(18.2)
z=1
where j is the reaction number, Nr is the number of reactants, Mi is the molecular weight of species
i. The symbol j describes the molar rate of reaction j with unit [mole cm3 s1 ] with i j the socalled stoichiometric coecient of species i in reaction j. The eect of the turbulent mixing model is
captured by the symbol Di .
z
It should be noted that each zone has its own temperature, volume and mass of species while the
pressure is same for all zones. The zones are coupled to each other mainly by the equation of state,
pVz = mz
Ru
Tz
Mz
(18.3)
342
where Ru is the universal gas constant and Mz is the zones average molar mass. To close the system
of equations nally the constraint that the summation of the zone volumes is equal to the total volume
at each time t yields:
Nz
Vz (t) = V(t).
(18.4)
i=1
Here V(t) is the total cylinder volume as function of time, prescribed by the engines geometry (with
a so-called crank-slider model) and its angular speed (equal to [360 RPM/60]) .
In the code the four most common heat loss models are implemented using according to
Qh = A(T a T w )
(18.5)
where is either constant or determined by the Nusselt-like correlations of Hohenberg, Woschni [31]
or the modied-Woschni correlation by Assanis [32]. T a is the average cylinder temperature dened
using the equation of state consistently
Nz
nRu T a =
nz Ru T z
i=1
with n, nz [mole m3 ] the total molar concentration and that of each zone respectively. In short, the
unknowns that need to be solved in the model are pressure, p, and for each zone temperature, volume
and mass of each species, viz. Vz , T z , mz,i . Since detailed chemical mechanisms are used, the set of
coupled non-linear equations is large (100-1500 equations per zone) and typically sti. Consequently,
an ecient and reliable solver has to be applied of which here the DASSL-code of Petzold is used
[33].
Chemical models
To do an actual simulation a model needs to be provided that describes all elementary reaction steps
relevant for the ignition and conversion of the fuel at hand. However, the chemistry of real automotive
fuels, like Diesel and Gasoline, is not totally comprehended due to the vast number of components
in it. As such the concept of surrogate fuels receives a lot of attention. Surrogate fuels are based on
a (necessarily) small set of components, like the primary reference fuels (PRFs), e.g. n-heptane and
iso-octane (others exist see [34]) augmented with additional sub mechanisms (e.g. toluene) to model
the real fuel chemical behaviour over a wider range of conditions8 .
In the CT-group at the TU/e two dierent chemical models are employed mainly. The rst one is
a semi-detailed mechanism gasoline surrogate (Andrae [35]) based on a skeletal n-heptane/iso-octane
PRF model to which toluene, di-isobutylene and ethanol mechanisms are added. To be able to study
the NOx formation, the NOx mechanism of GRI3.0 is added resulting in a mechanism consisting of
153 species and 723 reactions. The second mechanism is a more detailed n-heptane mechanism of
Seiser et al. [36]. It is based on the detailed n-heptane mechanism of Curran et al. [37] (556 species
and 2540 reactions) reduced systematically to 179 species and 896 reactions. Here the same NOx
mechanism is added as well.
8
Note that if only n-heptane and i-octane would be used the RON and MON number of the blend would be identical.
Regular fuels always have dierent RON and MON numbers. For that reason Toluene, a cyclic molecule, is blended in.
Other choices (Xylene) are used as well in literature.
343
Initialization
To initialize a simulation, the air/EGR composition, temperature and volume of each zone needs to
specied. These are retrieved from the experimental data directly, see cf. [38]. The stratication,
however, can not be obtained from these experiments as it is not or put more accurately cannot be
measured in a full metal engine. At the TU/e CFD models are applied developed in StarCD, cf. [39].
It is used to simulate the evolution of the fuel distribution as a function of the injection timing. As an
example in gure 18.8 the computed fuel distribution for a SOI=50o bTDC case is presented.
Figure 18.8: Sample fuel fraction contours at dierent crank angles (40o , 25o and 15o bTDC, respectively)
To determine dierent levels of stratication in the multi-zone simulations the actual fuel distribution per simulated case is sampled at various crank angles (gure 18.9). The corresponding amount
of fuel is then injected in each zone to obtain the equivalent mixture stratication.
0.14
3 deg aEOI
8 deg aEOI
13 deg aEOI
18 deg aEOI
0.12
Fuel Fraction
0.1
0.08
0.06
0.04
0.02
4
6
Zone Number
10
Figure 18.9: Fuel fraction distribution as a function of volume for dierent sample angles.
344
18.4.2
Typical results
HCCI
If initialized with one zone the result will be the true HCCI limit. As Diesel engines operate with
overall lean mixtures ( 1.25, 0.8 typically) this will not give rise to a signicant CO and UHC
level if the simulation does not fail to ignite. For a typical case (40% EGR, 4 bar IMEP) in gure 18.10
the pressure and temperature is given. As a reference the same simulation but now computed with air
instead of EGR is included as well. The typical two-stage ignition process (arrows in sub b) and c)),
which is common for higher hydrocarbons, can be observed in all gures.
Figure 18.10: Overview of simulation results. Solid lines:40% EGR, dotted lines:Pure Air results.
The application of EGR indeed results in later ignition ( 3CA) due to the reduced temperature
because of the higher heat capacity of CO2 and H2 O. Maximum temperature and pressure are also
lower and consequently this should yield lower NOx values9 (1 [ppm] vs 3 [ppm]!). This is indeed the
case as can bee seen in gure 18.11. Also nicely illustrated here is that there is no discernable emission
of all the other species CO, UHCandSoot. CO, UHC is simply due to the fact that combustion if started
in time will always lead to complete combustion in these lean conditions ( = 0.3 and = 0.47
respectively).
PCCI
Just for illustrative purposes the development of the zonal volume in a 5-zone simulation is presented.
The simulation is initialized using the CFD model(SOI=45oCA, stratication sampled at 25oCA,
40% EGR). From the right gure it can be observed that between -30 oCA and -20 oCA the volume
9
CA!
Emission levels of an engine are determined by the concentrations at exhaust valve open (EVO), which is towards 180
345
Figure 18.11: Main emissions. Solid lines: 40% results, dotted lines: Pure Air results.
Figure 18.12: The development of the zonal volume as function of time(solid lines). The dotted line equals
Va = V(t)/Nz . On the right the dierence Vz Va is presented.
of each zone starts to deviate from the average. This is because of evaporative cooling. As each
zone receives a dierent amount of fuel it will experience a dierent amount of cooling. Clearly the
richest zone (zone 5) becomes the smallest zone just prior to its ignition. After the point of ignition
the dynamics become more complicated. For instance zone 4 ignites rst thereby compressing the
others. At some point the others ignite but due their dierent fuel content expand less but for zone
5. As mentioned earlier zone 5 contains the most fuel and consequently shows the largest increase in
volume in the end .
Interesting is now to see if the stratication has impact on the emissions. As can be observed in
gure 18.13 stratication has a large impact on all emissions, changing it by orders of magnitude.
NOx -formation is mainly aected by temperature. Apparently the zonal temperatures in the PCCI
346
Figure 18.13: Main Emissions. Single zone (HCCI) vs. multi-zone (PCCI). The numbers in the legend of each
subplot indicate the level of emissions at EVO in ppm.
case exceed the temperature of the HCCI-case as is indeed the case in gure 18.14). From the right
plot in the same gure it shows that up to 70% of the total comes from the richest and hottest zone.
The fact that all zones approach a common value at the end is due to the inter-zonal mixing model.
Figure 18.14: Left: Temperatures. Right: NOx . Single zone (HCCI) vs. multi-zone (PCCI).
18.4.3
Outlook
On one hand the numerical research at the TU/e will develop further along the multi-zone methodology as it allows for the use of detailed mechanisms without the need for a turbulent closure. Clearly
more eort needs to be spend on mixing models and chemical mechanisms involving even larger
347
components. The code will be used to test and develop so-called surrogate fuels that mimic realworld fossil-based fuels by only a few yet representative hydro-carbons. Performance of bio-fuels can
be studied. For simple bio-fuels like alcohols, DME etc., adequate chemical mechanisms exist. For
newer and more complex bio-fuels mechanism might need to be developed.
On the other hand an extension of the FGM methodology to model HCCI/PCCI combustion in a
CFD setting is currently developed. Although a closure for the combustion model is needed the eect
of turbulent mixing and geometrical eects can be studied more easily.
348
Appendix
Boschs rst line-pump
Figure 18.15: (One o) the rst high pressure fuel pump. Drawing from a 1934 patent [1944858]. The principal
design of Diesel FIE hasnt changed up to the late 90s when common-rail systems emerged.
Measurement matrix
Over all experiments, a number of parameters are chosen to be kept constant over the whole series.
To investigate the eects of several parameters on PCCI combustion, rst the in cylinder conditions
are varied using European diesel as fuel. In the second part of this measurement series, the operating conditions (table 18.2) resulting in the most ecient combustion, both in fuel consumption and
emissions, are also tested for other fuels. Results of the complete study can be found in [43] and [44].
Heat capacity of EGR/AIR mixtures.

Figure 18.16 shows the isentropic coecient ( = c p /cV ) for certain EGR/AIR mixtures as a function
of temperature.
349
Table 18.2: Operating Conditions
Engine Speed
Fuel injection pressure
Fuel temperature
Intake air temperature
Injection durations
Intake pressure levels
EGR levels
1200 rpm
1500 bar
307 K
300 K
900 - 1200 - 1600 s
1.25 - 1.5 bar
0 - 15 - 40 - 50 - 60 wt
1.4
[]
1.38
0.0
0.2
0.6
EGR fraction
1.36
1.34
1.32
1.3
1.28
1.26
0
500
1000
1500
T [K]
2000
2500
3000
Figure 18.16: Isentropic coecient ( = c p /cV ) for certain EGR/AIR mixtures as a function of temperature.
Evaluated using NASA polynomials.
References
[1] Akihama, K., Takatori, Y., Inagaki, K., Sasaki, S., and Dean, A.M., Mechanism of the Smokeless
Rich Diesel Combustion by Reducing Temperature, SAE paper 2001-01-0655, (2001).
[2] Will F. Colban, Duksang Kim, Paul C. Miles, Seungmook Oh, Richard Opat, Roger Krieger and
David Foster, Russell P. Durrett, Manuel A. Gonzalez D, A Detailed Comparison of Emissions
and Combustion Performance Between Optical and Metal Single-Cylinder Diesel Engines at
Low Temperature Combustion Conditions, SAE paper 2008-01-1066, (2008)
[3] Caroline L. Genzale, Rolf D. Reitz and Mark P. B. Musculus, Eects of Piston Bowl Geometry on
Mixture Development and Late-Injection Low-Temperature Combustion in a Heavy-Duty Diesel
Engine, SAE paper 2008-01-1330, (2008)
350
[4] Will F. Colban, Paul C. Miles, Seungmook Oh, Eect of Intake Pressure on Performance and
Emissions in an Automotive Diesel Engine Operating in Low Temperature Combustion Regimes,
SAE paper 2007-01-4063, (2007)
[5] Raj Kumar, Ming Zheng, Usman Asad and Graham T. Reader, Heat Release Based Adaptive
Control to Improve Low Temperature Diesel Engine Combustion, SAE paper 2007-01-0771,
(2007)
[6] Satbir Singh, Rolf D. Reitz, Mark P. B. Musculus, Comparison of the Characteristic Time (CTC),
Representative Interactive Flamelet (RIF), and Direct Integration with Detailed Chemistry Combustion Models against Optical Diagnostic Data for Multi-Mode Combustion in a Heavy-Duty
DI Diesel Engine, SAE paper 2006-01-0055, (2007)
[7] Sanghoon Kook, Choongsik Bae, Paul C. Miles, Dae Choi, Michael Bergin and Rolf D. Reitz,
The Eect of Swirl Ratio and Fuel Injection Parameters on CO Emission and Fuel Conversion
Eciency for High-Dilution, Low-Temperature Combustion in an Automotive Diesel Engine,
SAE paper 2006-01-0197, (2006)
[8] Cherian A. Idicheria and Lyle M. Pickett, Soot Formation in Diesel Combustion under HighEGR Conditions Sandia National Laboratories, SAE paper 2005-01-3834, (2005)
[9] S. Delhaye, L.M.T. Somers, J.A. van Oijen, L.P.H. de Goey, Incorporating unsteady ow-eects
in Flamelet-Generated Manifolds, accepted in Combust. Flame, (2008)
[10] S. Delhaye, L.M.T. Somers, J.A. van Oijen, L.P.H. de Goey ,Incorporating unsteady ow-eects
beyond the extinction limit in Flamelet-Generated Manifolds, accepted for publication in 32nd
International Symposium of the Combustion Institute, (2008)
[11] Ryo Hasegawa and Hiromichi Yanagihara, HCCI combustion in DI diesel engine, SAE paper
2003-01-0745, (2003)
[12] Gautam T. Kalghatgi, Per Risberg and Hans-Erik ngstr m, Partially pre-mixed auto-ignition
o
of gasoline to attain low smoke and low NOx at high load in a compression ignition engine and
comparison with a diesel fuel, SAE paper 2007-01-006, (2007)
[13] Gautam T. Kalghatgi, Per Risberg and Hans-Erik ngstr m, Advantages of fuels with high reo
sistance to auto-ignition in late-injection, low-temperature, compression ignition Combustion,
SAE paper 2006-01-3385, (2006)
[14] Tomohiro Kanda, Takazo Hakozaki, Tatsuya Uchimoto, Jyunichi Hatano, Naoto Kitayama and
Hiroshi Sono, PCCI Operation with Early Injection of Conventional Diesel Fuel, SAE 2005-010378, (2005)
[15] Tomohiro Kanda, Takazo Hakozaki, Tatsuya Uchimoto, Jyunichi Hatano, Naoto Kitayama and
Hiroshi Sono, PCCI Operation with Fuel Injection Timing Set Close to TDC, SAE paper 200601-0920, (2006)
[16] Shuji Kimura, Osamu Aoki, Yasuhisa Kitahara and Eiji Aiyoshizawa, Ultra-Clean Combustion
Technology Combining a Low-Temperature and Premixed Combustion Concept for Meeting Future Emission Standards, SAE paper 2001-01-0200, (2001)
351
[17] Shuji Kimura, Osamu Aoki, Hiroshi Ogawa and Shigeo Muranaka, Yoshiteru Enomoto,New
Combustion Concept for Ultra-Clean and High-Eciency Small DI Diesel Engines,SAE paper
1999-01-3681, (1999)
[18] E.Doosje, Limits of mixture dilution in gas engines, PhD Thesis, Technische Universiteit Eindhoven, (2008) in press
[19] P.F.M. Frijters, PhD Thesis , Technische Universiteit Eindhoven, (2011)
[20] M.D. Boot, P.J.M. Frijters, R.J.H. Klein-Douwel, R.S.G. Baert, Oxygenated fuel composition
impact on heavy-duty Diesel engine emissions, SAE Technical Papers, -, 2007-01-2018, (2007)
[21] P.J.M. Frijters, R.S.G. Baert, Oxygenated fuels for clean heavy-duty diesel engines, Int. J. of
Vehicle Design, 41(1/2/3), 242-255, (2006)
[22] Bohm,B., Kittler,C. and Dreizler, A., Diagnostics at high repetition rates: new insights into
transient combustion phenomena, Proceedings of the 3rd European Combustion Meeting,5-1,
(2007)
[23] M. Boileau ,G. Staelbach, B. Cuenot, T. Poinsot, C. B rat,LES of an ignition sequence in a gas
e
turbine engine, Comb. Flame 154, 2, (2008)
[24] R.J.H. Klein-Douwel, A.J. Donkerbroek, A.P. van Vliet, M.D. Boot, L.M.T. Somers, R.S.G.
Baert, N.J. Dam, J.J. ter Meulen, Soot and chemiluminescence in diesel combustion of bioderived, oxygenated and reference fuels, Proc. Combust. Inst., 32, 1-17, (2008)
[25] M.D. Boot, P.F.M. Frijters, C.Luijten, L.M.T. Somers, R.S.G. Baert, A.J. Donkerbroek, R.J.H.
Klein-Douwel, N.J. Dam, Cyclic Oxygenates: A New Class of Second Generation Biofuels for
Diesel Engines, Energy and Fuels, (2008)
[26] J.E. Dec,A conceptual model of DI diesel combustion based on laser-sheet imaging,SAE paper
970873, (1997)
[27] http://en.wikipedia.org/wiki/Robert Bosch
[28] http://en.wikipedia.org/wiki/Rudolf Diesel
[29] Dec, J.E. Advanced Compression-Ignition Engines - Understanding the In-Cylinder Processes,
Proceedings of the Combustion Institute, Vol. 32 ,pages 2727-2742,(2009)
[30] Aceves, S. M. and Flowers, D. L., A Detailed Chemical Kinetic Analysis of Low Temperature
Non-Sooting Diesel Combustion, SAE Technical Papers, 2005-01-0923, (2005)
[31] Woschni, G. A universally applicable equation for the instantaneous heat transfer coecient in
the internal combustion engine, SAE 670931, (1967)
[32] Soyhan, H.S., Yasar, H., Walmsley, H., Head, B., Kalghatgi G.T., Sorusbay, C., Evaluation of
heat transfer correlations for HCCI engine modeling, Applied Thermal Engineering, Volume
29, Issues 2-3, Pages 541-549, (2009).
[33] Petzold, L.R. A description of dassl: A dierential/algebraic system solver, Scientic Computing, page 65-68, (1983)
352
[34] Farrell, J.T., Cernansky, N.P., Dryer, F.L., Friend, D.G., Hergart, C.A., Law, C.K., McDavid,
R.M., Mueller, C.J., Patel, A.K. and Pitsch, H. Development of an Experimental Database and
Kinetic Models for Surrogate Diesel Fuels, SAE Technical Paper 2007-01-0201,(2007)
[35] Andrae, J.C.G., Brinck T. and Kalghatgi, G.T. HCCI experiments with toluene reference fuels
modeled by a semi-detailed chemical kinetic model, Combustion and Flame, 155, pages 696-712,
(2008)
[36] Seiser, R., Pitsch, H., Seshadri, K., Pitz, W.J. and Curran, H.J. Extinction and Autoignition of
n-Heptane in Counterow Conguration, Proc. Comb. Inst, 28, 2029-2037, (2000)
[37] Curran, H. J., Gauri, P., Pitz, W. J. and Westbrook, C. K., A Comprehensive Modeling Study of
n-Heptane Oxidation , Combust. Flame 114:149-177, (1998)
[38] Eg z, U. Numerical Investigation of Mixture Formation and Combustion in PCCI conditions,
u
Master Thesis, Eindhoven University of Technology, Combustion Technology, (2009).
[39] Rijk, E. CFD Modeling of Fuel Injection and Combustion in an HDDI Engine, Master Thesis,
Eindhoven University of Technology, Combustion Technology, (2009)
[40] Boot, M.D., et al. Optimization of Operating Conditions in the Early Direct Injection Premixed
Charge Compression Ignition Regime., SAE Technical Paper, 2009-24-0048, (2009)
[41] Boot, M.D., et al. Uncooled External EGR as a Means of Limiting Wall-Wetting under Early
Direct Injection Conditions, SAE Technical Paper 2009-01-0665, (2009)
[42] Frijters, P.J.M., Implementatie Data Acquisitie en Functionaliteitstest van een DAF Heavy Duty
Dieselmotor. s.l.: Technische Universiteit Eindhoven, TU/e report no. WVT 2002.07, (2002).
[43] Leermakers, C.A.J. Experimental Study on the Impact of Operating Conditions and Fuel Composition on PCCI Combustion, Master Thesis, Eindhoven University of Technology, Combustion
Technology, (2010).
[44] Van den Berge, B. Experimental Study on the Impact of Fuels with Lower Cetane Number on
PCCI Combustion, Internal Report, Eindhoven University of Technology, Combustion Technology, (2010).
[45] Boot, M.D. Approaches to Improve Mixing in Compression Ignition Engines, PhD Thesis, Eindhoven University of Technology, Combustion Technology, (2010).
Chapter 19
Spray combustion in engines

C.C.M. Luijten
Obviously, a one-hour lecture and some pages of lecture notes are by far not sucient to cover the
broad subject of spray combustion in engines (which can be characterized as non-premixed, nonequilibrium, two-phase, non-homogeneous and turbulent). As a matter of fact, dedicated journals
exist on sprays; even more so is true for combustion in engines, and for fuel composition eects in
engine combustion. Therefore, the current lecture can not pretend to be more than an introduction.
Still, by providing ample references, the reader should be able to nd his/her way into the literature
with the help of the lecture slides and the text and references in this chapter. Optimally, this should
function as a Quick Start Guide to the vast literature on spray combustion in engines.
The scope of this chapter is as follows. After an introductory section, in which the relation with the
rest of the course is pointed out, we will look into the phenomenology of fuel sprays (e.g., with what
speed and angle they penetrate the combustion chamber, how they break up, evaporate and ultimately
combust). After that we will briey look into the dierent classes of spray models, ranging from simple
measurement-based correlations to full CFD models (without going into details on the latter). The
core of this chapter, section 19.4, is devoted to (semi-)phenomenological modeling of spray processes,
since in my opinion such models provide a good balance between simplicity and physical insight.
19.1
Introduction
From the previous chapter it is clear that fuel-air mixing is of utmost importance for the eciency and
emissions of the combustion process in engines. These days, fuel is very often introduced into the
Internal Combustion Engine in the form of a spray. This holds true both for SI (spark ignition) and CI
(compression ignition) engines; however, the spray regimes dier greatly.
In SI engines, the fuel is introduced together with the intake air, implying that injection takes
place in a low temperature, low density environment. This has no great impact on the fuel-air mixing
process: fuel volatility is generally high, and due to the early injection there is ample time for mixing.
As a result, fuel and air in an SI engine are generally well premixed. Only few modern SI engines
are using the concept of DI (direct injection), in which fuel is injected during the compression stroke.
The result is then an overall lean mixture, which is however slightly rich near the spark plug, to
facilitate spark ignition. This implies that some stratication is present, which is needed for optimal
combustion in such (SIDI) engines.
Notwithstanding the importance of DI developments in SI engines, the current text (and lecture)
will mostly focus on the direct injection process in DI diesel engines. Due to their generally higher
compression ratio, combined with a broader use of turbochargers, the density at the moment of fuel
353
354
Chapter 19 C.C.M. Luijten

Thermal NO Production Zone
Liquid Fuel
Diffusion Flame
Rich Fuel/Air Mixture

Fuel-Rich Premixed Combustion
Soot Oxidation Zone
Initial Soot Formation
10
Scale (mm)
20
Low
High
Soot Concentration
Figure 19.1: Conceptual model of quasi-steady reacting diesel spray, adapted from Dec [14]
injection is generally much higher. The same holds to some extent for temperature. As we will see,
this has important implications for the fuel evaporation process in the spray.
In conventional diesel combustion, only a small portion of the fuel would burn in premixed mode.
This holds for the portion of fuel that is injected rst: it needs some time to vaporize, heat up and
auto-ignite (i.e. the ignition delay, ID). After the ID, the resulting premixed region combusts quickly,
resulting in the so-called premixed peak in the heat release, associated with the characteristic sound
of a diesel engine (diesel knock).
The remainder of the fuel is then injected into this rich, premixed ame; as a result, most of
it necessarily burns in a diusion mode at the periphery of the spray. The process just sketched is
nowadays understood fairly well; much of this understanding is often attributed to John Dec, who
published his conceptual model on diesel combustion in 1997 [14]. The main summarizing picture
from that reference is reproduced in Fig. 19.1. The picture shows a mature spray, which has reached
its quasi-steady state (i.e. the upstream part of the spray stays more or less the same, whereas the head
vortex is penetrating the combustion chamber further) after the injection transient. The latter is also
addressed in Ref. [14], as well as in the lecture slides.
Recently, more and more emphasis is put on premixed combustion modes for diesel engines as
well, as sketched in the previous chapter. In such combustion modes, ignition delay is generally larger,
aiming at a separation of the injection and combustion events. Obviously, the resulting stratication
of the fuel-air mixture (and, not unimportantly, of temperature!) prior to combustion is crucial to the
eciency and emissions reduction of these new combustion processes and, hence, of future engines.
In order to understand from basic principles how spray formation takes place, and how it results
in a certain distribution of fuel/air ratio and temperature, this chapter deals with spray processes in
a structured way. The description focuses on mixture formation before combustion, since a detailed
understanding of the combustion process is necessarily beyond the scope of a one-hour lecture.
Hence, this chapter focuses on physical spray processes, including dynamics (spray penetration,
breakup, air entrainment) and thermodynamics (evaporation). A smaller part of the chapter is devoted to the actual combustion phase. A better name for the chapter might therefore be sprays in
combustion engines rather than spray combustion in engines.

19.1.1
355
Life of a fuel spray
According to Wikipedia, a spray is a dynamic collection of liquid drops and the entrained surrounding
gas. To my taste, this denition is a bit (too) broad, since it may also apply to, for instance, a rain
cloud. Consulting some more sources on the web, it appears that in spite of a very broad range of
applications of sprays there are common properties applying to all sprays.
Indeed a spray consists of droplets and entrained ambient gas. The existence of these droplets
is the result of atomization, in which a liquid jet breaks up into droplets due to its high velocity.
Typically, a spray issues from a small hole connected to a pressurized container. The liquid jet leaves
the hole with very high velocity, giving rise to high levels of shear with the surrounding gas. In
addition, a high speed liquid jet also exhibits intrinsic instabilities (some of these are briey addressed
later on), which present an additional cause for breakup.
In any case, the result is a ne mist of liquid droplets, which protrudes into the surrounding gas,
entraining a signicant portion of it. In this way, momentum is transferred from the droplets to the
gas; as a result, the spray velocity decreases with distance from the nozzle hole. This process can be
modeled with high accuracy using fairly simple models, as will be shown in section 19.4.
The tip of the spray travels into the gas in the form of a head vortex, which typically makes out
20 to 30% of the size of the full spray. The region behind the head vortex shows more or less straight
edges (that is, on average; turbulent uctuations are always present). Most air entrainment takes place
along these edges.
The above description holds for all sprays, not only for those used in combustion systems. Typical
for the latter (fuel sprays) is that the surrounding gas is hot. In addition, especially for DI diesel
engines, the fuel injection pressure is typically very large (several thousand of bars), giving rise to
extremely high velocities. Hence, the droplets resulting from atomization are extremely small, and
quickly vaporize in the hot ambient gas.
Since combustion requires mixing on a molecular level, it only sets in after sucient evaporation.
Therefore, part of the observed ignition delay (ID) in engines is due to the evaporation process (the
so-called physical ignition delay). The other part is due to the auto-ignition chemistry of the fuel
(chemical ignition delay). It should be noted that these phases do not strictly follow each other,
but take place simultaneously (although evaporation temperatures might be too low for signicant
auto-ignition chemistry to occur).
During the ignition delay period, the rst portion of the spray has had time to premix with air.
As pointed out before, this part then quickly combusts in a premixed rich ame. The trailing edge of
the spray enters this hot premixed ame and also ignites. The heat of combustion aects evaporation;
likewise, the evaporation process will aect combustion (via temperature) due to evaporative cooling.
After some time, a quasi-steady equilibrium distance sets in where the foot of the turbulent spray
ame is located, the so-called ame lift-o length H (the precise stabilization mechanism is still not
fully understood, as will be addressed later).
All fuel entering this premixed ame will, at rst, partly combust; the rich products travel further
downstream, giving rise to soot formation. Ultimately, these rich products and most of the soot
will oxidize in the turbulent diusion ame surrounding the spray. This very hot diusion ame front
is the main location of NO formation in the engine. The resulting heat release causes the pressure to
increase, which ultimately drives the engine. Unfortunately, part of the soot may not oxidize and is
emitted with the exhaust gas.
When injection stops, the existing ame will consume the fuel remaining upstream. However, in
cases where the ame lift-o is large, this part may become too lean to combust, in which case the
end phase of injection is a source of unburnt hydrocarbons [50, 61, 47].
356
19.1.2

Measurement in fuel sprays
From the previous section it appears that a fuel spray has a fairly complex life from the cradle to
the grave (end of combustion). Many processes take place, partly sequential, but mostly in parallel.
Some of these processes are strictly physical in nature, others are (also) chemical. To understand
these processes, how they interact, and how they ultimately translate into engine performance (power,
eciency) and emissions, requires a wide range of well-designed experiments.
Such experiments are complicated by the fact that an engine is not the easiest environment for
measurements, to say the least. Pressures and temperatures are very high; the environment is highly
reactive; and the geometry is constantly moving at a very high speed. To enable this combination
of conditions, engines are necessarily constructed out of metal parts. Unfortunately, metals are not
transparent. Consequently, most engine experiments are limited to measuring input-output relations
(inputs: speed, intake pressure and temperature, fueling etc.; outputs: power, eciency, emissions).
Often, in-cylinder pressures are also monitored to enable some analysis of the in-cylinder processes.
All these measurements are indirect and do not exhibit much detail of the spray process.
In order to capture the behavior of a combusting fuel spray, optical measurements are necessary.
Introducing optical access to an engine is far from easy, although it can be done. Unfortunately, the
engine will be dierent after such modication, depending on the extent of optical access provided:
think of a changing piston head geometry (when replacing it with a typically at window), and
changing heat transfer rates for a Bowditch engine principle (where the piston head is put on legs).
In spite of these drawbacks, visualization of spray combustion in real engines has been done by several groups (including the Nijmegen Molecular Physics group and the Eindhoven CT group). Some
example results are shown in the lecture.
Measurements in optical engines are often not very quantitative. For a detailed understanding of
fuel sprays and validation of spray models, measurements must be done in dedicated test setups. In
such setups the engine environment is mimicked (providing high pressure and temperature), but with
better optical access. Examples include shock tubes, rapid compression machines, modied (motored)
engines, and so-called constant volume cells or combustion bombs. An overview together with
pros and cons of each is provided in a recent SAE paper by Baert et al. [7].
At Eindhoven University of Technology, spray research is mainly done in the Eindhoven High
Pressure Cell (EHPC). Its operating principle is described at length in the paper just mentioned, and is
briey summarized here. A schematic picture is provided in Fig. 19.2. The EHPC is a constant volume
cell, in which high pressure and temperature are created using pre-combustion. In this method, the
EHPC is lled with a mixture of acetylene, nitrogen, oxygen and argon. The mixture is ignited by a
spark and then explodes; when combustion is complete, the end gas cools down and then a fuel spray
is injected with a chosen timing, based on the average temperature at that moment in the cell. The
ambient gas density is constant over time, governed by the amount of mixture that is initially entered.
Depending on the initial gas composition, the injected fuel spray may or may not react. When
pre-combustion is close to stoichiometric, no oxygen is left and the fuel spray will only vaporize. In
the other case, when pre-combustion starts with a lean mixture, oxygen is left after it, and the fuel
spray may still react. The properties of the pre-combustion end gas can be tuned to desired values
(density, peak temperature, remaining oxygen fraction and specic heat) by adjusting the four intake
masses (acetylene, oxygen, nitrogen, argon), as discussed in Ref. [7].
Since a constant volume cell has no moving parts, optical access can be granted easily through
large windows (both quartz and sapphire windows are used; quartz has a better transparency for UV
wavelengths, but sapphire can handle higher pressures). In Ref. [7], several examples of optical
measurements are discussed. So far, mostly macroscopic spray phenomena (penetration, cone angle,
357
liquid length, ame lift-o) have been measured. This was mostly done using a combination of
shadowgraphy, Schlieren, Mie scattering and natural luminosity measurements. Sample results are
shown in the lecture. More generally, it must be noted that a variety of laser diagnostic techniques has
contributed enormously to the understanding of sprays, both before [76] and during combustion [14].
For reasons of time and space, these methods will not further be discussed here.
Figure 19.2: Cross-sectional overview of the Eindhoven High Pressure Cell and its main components [7]
So far we did not address fuel injection equipment. For an excellent introduction, the reader is
referred to Baumgarten [8]. For DI diesel engines, common rail systems are now most widely used.
In such a system, multiple injectors are connected to a buer vessel that is continuously kept on high
pressure in contrast to a conventional line pump, which works intermittently. The injector itself is
mostly operated by electronically actuating the return valve, which causes the needle to rise. As a
result, fuel is admitted to the nozzle holes (typically 100 to 200 m in diameter).
Important for the spray formation process are mainly the mass ow and momentum ow out of
an injector. To characterize its mass ow, the simplest way is collecting a number of injections in a
container, weighing it, and dividing by the number of injections. In this way the integrated mass ow
per injection is obtained. Comparing it with the theoretical value (see next section) provides a value
of the discharge coecient Cd .
If time-resolved mass ow results are needed, the injector can be mounted into a so-called Zeuch
chamber. This is a vessel that is pre-lled with the fuel of interest. Adding extra fuel gives rise to a
strong pressure increase, caused by the incompressibility of the liquid fuel. By measuring the time
resolved pressure, the mass ow during an injection can be obtained, which is important for engine
calibration. More details on this method are provided by Seykens et al. [70].
To characterize the momentum ow from a nozzle, one should realize that force is equal to the
change of momentum. Holding a plate perpendicularly in front of an injector nozzle hole, will cause
the fuel spray to loose its momentum (since the fuel can only move perpendicularly to the plate).
Connecting the plate to a force transducer, and thus measuring the associated force, gives a measure
of the momentum ux from the nozzle hole. Although this measurement could in principle be done
in a time-resolved fashion, usually only the average force is considered, and is then used to derive the
momentum coecient C M (further addressed below), cf. Ref. [59].
358
19.2
Spray phenomenology
After the brief overview of the life of a spray in the preceding section, this section goes into more
detail on the main processes involved in spray combustion. For each of these processes, the main
parameters are introduced. By making the reader familiar with these concepts, it is hoped that the vast
literature on spray combustion becomes better accessible. A number of references is provided as a
starting point (but note that the cited literature is by no means intended to be complete).
19.2.1
Nozzle discharge
The ow through an injector nozzle hole is mainly governed by Bernouillis law. The theoretical exit
velocity Uth and the pressure dierence over the nozzle hole p are coupled according to
Uth =
2p
,
f
(19.1)
where f is the (liquid) fuel density. The resulting (theoretical) mass ow rate equals mth = f Uth A0 ,
2 /4 is the geometrical nozzle cross-sectional area. Now Bernouillis law is only valid
where A0 = d0
for an incompressible, frictionless ow, which is not fully true during nozzle discharge. However, the
dierence can be repaired by relatively simple corrections.
What actually happens in a nozzle hole is depicted in Fig. 19.3, taken from Payri et al. [58]. Fuel
from the pressure reservoir rounds the inner corner, and strongly accelerates. Due to the resulting
high velocities, the dynamic pressure strongly drops, causing local pressure to drop below the fuels
saturation pressure. Consequently, the ow is cavitating. The bubbles formed are partly transported
downstream with the fuel, changing the average density of the ow, as depicted on the right hand side
of Fig 19.3. This can be interpreted as if the ow would have an eectively smaller outow area for
fuel of (liquid-only) density l . The ratio of this eective area and the geometric area is Ca = Ae f f /A0 ,
known as the area contraction coecient.
The assumption of frictionless ow is also not fully correct: the exit velocity is smaller than
the value obtained from Eq. (19.1). The dierence is expressed in terms of the velocity contraction
coecient, Cv = Ue f f /Uth . Combining both eects, the real mass ow from the nozzle hole can be
written as m = f Ue f f Ae f f . We can then dene the ratio between the actual and theoretical mass
ow; this is the discharge coecient Cd , dened by

Cd =
U e f f Ae f f
m
= Cv Ca .
mth
Uth
A0
(19.2)
Following similar arguments, we can also compare the real momentum ux with the theoretical one.
2
2
The latter obviously is Mth = f Uth A0 ; similarly, the eective momentum ux is M = f Ue f f Ae f f .

Comparing the two in a ratio provides the momentum coecient C M , given by
CM =
Ue f f
M
=
Uth
Mth
Ae f f
2
= Cv Ca = Cd Cv .
A0
(19.3)
Eectively we have chosen Ae f f and Ue f f such, that the resulting mass and momentum ows follow
the idealized expressions above. As indicated in the previous section, both Cd and C M can be obtained
from fairly straightforward experiments. Combining them, using the preceding two equations, enables
one to obtain the velocity and area contraction coecients as well, as detailed for instance in Ref. [58].
These values are needed in most of the phenomenological models to be discussed in section 19.4.
359
Figure 19.3: Left: schematic overview of internal nozzle ow with ow separation. Right: denition of
eective velocity and surface area. Source: Payri et al. [58]
Unfortunately, the values of the constants thus obtained are not really constant with conditions.
They are found to depend on the amount of cavitation that occurs in the nozzle hole [58]. Cavitation
occurs if the pressure in point c in Fig. 19.3 drops below the fuels saturated vapor pressure pv . If the
ow accelerates even further, the thermodynamic equilibrium that sets in will keep the local pressure
equal to pv . Using Bernouilli from point i to c, continuity in c and the denition of Cd , it can be shown
cf. Desantes et al. [20] that the discharge coecient then equals
Cd = C c
pi pv
C c K.
pi pb
(19.4)
In this equation the cavitation number K is introduced, which is equal to 1 for strongly cavitating
ows. The larger K gets, the less cavitation occurs, until a value K = K crit is reached where cavitation
disappears (typically K crit 1.1). The constant C c = Ac /A0 is the ratio of the vena contracta surface
area to that of the nozzle hole, and is assumed to be constant for a given nozzle.
Coming from the other limit (i.e. initially non-cavitating ow at moderate values of pi ), K initially
is a (weak) function of the Reynolds number. When it gets below K crit , thus when cavitation sets in,
an interesting combination of phenomena happens, see Fig. 19.4. The momentum coecient C M is
hardly aected by cavitation and remains constant. The mass ow, however, is choked at the critical
value; a further decrease of K (for instance by lowering pb ) will not further increase m f . Accordingly,
Cd goes down with decreasing K. This can only happen see Eqs. (19.2) and (19.3) if Cv increases,
whereas Ca decreases more strongly. Physically, this is attributed to the eect of cavitation as follows:
the bubbles in the ow decrease the amount of wall shear, enabling the velocity to increase, but at the
same time decreasing the eective area of the liquid phase [58].
In diesel applications, back pressures are typically between 2 and 7 MPa; common rail pressures
are typically around 200 MPa. As a result, practical values of K are around 1.02, so very close
360
Figure 19.4: Left: qualitative behavior of discharge coecient Cd with cavitation number K.
Right: experimental data, also including C M and (derived) Ca , Cv values. Source: Payri et al. [58]
to 1 and in the sensitive area of Fig. 19.4. In principle, the above dependence of nozzle coecients on
conditions should be taken into account when using experimentally measured constants in applications
with dierent K; still, the variation of K is limited in an absolute sense, so as a rst approximation
using constant values is not that bad.
Apart from pressure conditions, discharge coecients depend on details of nozzle hole geometry.
Such eects on spray dynamics as well as on engine performance have been extensively studied
by the Valencia group, see for instance Refs. [57, 58, 59, 20]. Interestingly, in some of these works the
experimental ndings are combined with the inspection of polymer moulds of the nozzle holes. As an
example, it was found that convergent nozzle holes show almost no tendency for cavitation, which is
attributed to the fact that the entrance corner of such nozzles is less sharp.
In summary, we have seen in this section that nozzle discharge is governed by Bernouillis law;
deviations from it are partly attributable to the vena contracta eect, typically in combination with
cavitation. The resulting correction factors for mass and momentum ow and for velocity and surface
area have been discussed. In rst approximation, these correction factors are constant for a given
nozzle when conditions do not vary too widely. However, they do depend on individual nozzle hole
geometry.
Finally, a remark is in place about the injection pressure pi used above. It was so far assumed
to be constant; in practice, the internal ow dynamics of an injector (and the upstream common rail
system) causes signicant variation in pi (typically 5%) due to liquid phase compressibility eects.
As a result, the mass ow from the injector also varies during injection. For the concepts discussed
here this constitutes no big problem; for accurate engine operation, however, it needs to be taken in to
account. For this reason, quite some work is often done to model the hydrodynamic behavior of the
injection system (including common rail). Dedicated software is available that allows such modeling,
for instance AMESIM (see, for example, Ref. [70]) and HYDSIM (cf. [65]).
19.2.2
Breakup, droplet size and cone angle
From the previous section it is clear that the fuel, emanating from a nozzle hole in a modern diesel
injector, exits with a very high velocity and generally contains cavitation bubbles. Both observations
give rise to a process named breakup. The latter is subdivided into primary and secondary breakup.
During primary breakup, the remaining part of the intact liquid core is torn apart into ligaments and
(relatively large) droplets. Secondary breakup is the process by which these droplets are broken up
361
into smaller ones. In a review paper of Smallwood and G lder [76], ample evidence is provided that
u
breakup in contemporary diesel sprays has fully taken place after only a few nozzle diameters.
Primary breakup is the consequence of three eects: imploding cavitation bubbles (since the
pressure in the cylinder pb will normally be larger than the saturated vapor pressure pv ); turbulence
within the liquid phase; and to a lesser extent, cf. Refs. [76, 8] aerodynamic forces acting on
the liquid surface [77]. The balance between these mechanisms is governed by the Reynolds number
2
Re = f U0 d/ f (ratio of inertia to viscous drag) and the Weber number We = f U0 d/ f (ratio of
inertial to surface forces).
Ohnesorge [53] showed that a convenient
classication of breakup regimes could be obtained in
terms of Re and the Ohnesorge number Oh = We/Re = f / f f d (ratio between viscous drag and
surface tension). Since both Re and Oh are formulated in terms of liquid properties only (whereas the
density of the ambient gas denitely aects breakup), Reitz and Bracco latter added the density ratio
g /l as a third parameter [67].
Depending on these numbers, dierent regimes of primary breakup exist. For very high Re and
We (i.e. high velocity, resulting in a dominant role for inertial forces), the main breakup regime is
atomization, characterized by a fast and vigorous breakup into small droplets, a conical shape of the
liquid core at the nozzle and a diverging spray angle. The precise mechanisms behind it are still not
fully revealed, in view of the diculty to experimentally access the near-nozzle region.
Still, some detailed models of primary breakup are available. For the atomization regime, a model
describing the dominant eects of turbulence and cavitation exists [77, 8], describing the redistribution of the sum of turbulent kinetic energy and cavitation implosion energy, into vapor-liquid surface
energy and a velocity component perpendicular to the spray axis. The latter is directly responsible for
the characteristic conical shape of atomizing sprays (further addressed below).
During secondary breakup, fuel droplets further disintegrate. This process is primarily governed
by the Weber number, since it results from the trade-o between inertial forces (trying to tear the
droplet apart) and surface tension (keeping it together). Again dierent regimes exist; for more details
the reader is referred to literature, for example Refs. [77, 8]. The same holds for the description of
interaction of droplets with each other and (possibly) with the cylinder wall.
In terms of spray combustion, the two most important results of the breakup process are the resulting
spray cone angle and the droplet size distribution. From the above, it will be clear that a detailed
prediction of these parameters from the basic underlying processes is by far not easy. Although
commercial CFD packages often have breakup submodels on board, these contain a huge number of
tunable parameters, each with their own uncertainty. What is more, as discussed in the next section,
these parameters are often tuned in CFD computations in order to compensate grid dependencies.
As a result, predicted results of both cone angle and droplet size are often not very reliable. In
addition, we have seen in the previous section that, for instance, the cavitation process is signicantly
aected by details of the nozzle geometry. In other words, the precise balance and magnitude of
individual breakup processes will depend on details of each individual nozzle hole. De facto this will
also hold for cone angles and droplet size distributions.
In view of these diculties, many experimental eorts have been done to provide correlations for
cone angles and droplet size distributions in terms of other parameters (fuel and ambient properties
and injection parameters). Depending on what has been varied, these correlations capture more or
fewer eects or parameter dependencies. Examples are provided below, rst for droplet size and then
for spray cone angles.
362
Droplet diameters in sprays are most conveniently expressed in terms of Sauter Mean Diameter
(SMD). It denes an average droplet diameter in terms of having the right surface to volume ratio,
which is of importance in droplet evaporation. In other words, SMD is proportional to the total volume
of all droplets divided by their total surface area: from simple geometry, SMD = 6V/A [8].
Hiroyasu et al. [32] have correlated many measurements of SMD for relevant diesel conditions
as a function of injection parameters (nozzle diameter and injection velocity), liquid surface tension
and viscosity and density of both ambient gas and liquid. They provide two dierent correlations for
SMD, one for incomplete and one for complete atomization. The nal result of their work is
f
SMD
= max 4.12Re0.12 We0.75
d0
a
0.54
f
a
0.18
, 0.38Re
0.25
We
0.32
f
a
0.37
f
a
0.47
(19.5)
which is valid over a very large range of parameters. Among these are injection pressures up to
90 MPa and nozzle diameters down to 0.2 mm, which is just not sucient to cover modern diesel
conditions. The rst entry between square brackets pertains to incomplete sprays (associated with
less violent injection conditions). The second entry describes completely atomizing sprays, which
will generally be more relevant for our purposes.
In the original paper [32] the notation of kinematic and dynamic viscosity is not used consequently,
which gives rise to confusion (on a side note, the Hiroyasu group are not very strict in providing
proper nomenclature). Still, a detailed study of the manipulations in the paper has shown that the
above equation is correct if is interpreted as dynamic viscosity. In their nal Fig. 18, a very good
correlation quality of the above equation with their data is shown.
More importantly, that same gure shows that the great majority of values of SMD/d0 is between
0.1 and 0.3. Closer investigation of Eq. (19.5), writing the denitions of Re and We in full, shows that
0.07 so hardly depends on d and with U 0.39 . The higher the exit velocity,
SMD/d0 scales with d0
0
the smaller the normalized value of SMD gets. From this we can conclude that even smaller values
of SMD/d0 will be obtained for modern diesel conditions (higher injection pressures and smaller
nozzles).
Therefore, as an indication, SMD 0.1d0 can be used. This is conrmed by measurements of
Desantes et al. [15], who use nozzles holes of 0.23 mm and nd typical SMD values between 15 and
25 m. They also compare the predictions of Ref. [32] to their experimental ndings and conclude
that they t very well, in spite of the fact that the injection conditions are beyond the original t range.
Although other SMD correlations are available in literature, Eq. (19.5) still is the most popular,
since it is based on a very large quantity of measurements. Examples of its use can be found in
Refs. [82, 40]. In these papers, the assumption is made that all droplets have the same diameter SMD.
Yet, in principle, the size distribution is also important in spray modeling. In some other cases, a lognormal distribution having the same SMD is therefore assumed. However, droplet sizes resulting from
(primary and secondary) breakup in modern diesel engines are so small that they evaporate almost
instantly, depending mainly on the rate of air entrainment. This so-called mixing limited vaporization
will be discussed later. Hence, droplet size distributions might be less important after all, and are
therefore not further discussed here.
The spray cone angle plays an essential role in the models to be discussed in section 19.4, since
it largely determines the air entrainment rate. The latter in turn determines both the rate of evaporation
in the spray and the resulting equivalence ratio eld throughout the spray. Therefore, the cone angle
is of key importance for the (eciency and emissions of the) combustion process.
363
A great variety of correlations for exist. An often used one is from Hiroyasu and Arai [31]:
[ ] = 83.5
L
d0
0.22
d0
ds
0.15
a
f
0.26
(19.6)
where L the nozzle hole length and d s the sack diameter. Physically, if L/D is small, cavitation bubbles
do not collapse inside the nozzle but are convected outwards; by collapsing there, they increase the
cone angle. An older correlation by Hiroyasu and coworkers [30] is
2
d0 a p 0.25
= 0.05
2
a
(19.7)
A correlation recommended by Heywood [27] but originating from Reitz and Bracco [66] reads
4
a
tan(/2) =
3.0 + 0.28(L/d0 ) f
0.5
3
6
(19.8)
Naber and Siebers [52] provide the following cone angle correlation for vaporizing sprays:
tan(/2) = 0.27
0.19
f
0.0043
a
0.5
(19.9)
Not all of the above correlations are in terms of the same parameters; depending on what specic
authors have varied, they cast their t into a specic form. Another (seemingly trivial) point of concern
is that the cone angle and the half angle (/2) are often interchanged, leading to factors of 2 appearing
and disappearing again in dierent citations. The message is that one should always be very careful
using correlations from others; preferably these should be checked against in-house data. A second,
but equally important, argument to use in-house data is the eect of individual nozzle geometry,
discussed before.
The denition of the correlation parameters is also very important. As an example, correlation
Eq. (19.7) was measured in cold gases, and results were correlated as a function of (cold) gas viscosity.
If the fuel spray enters a hot cylinder, should viscosity then be determined at in-cylinder conditions
or at standard conditions? The answer can only be provided by comparison with hot measurements in
which again viscosity is varied however, such measurements are often not available.
In spite of the above remarks, we can conclude that the spray cone angle in all cases depends on
the density ratio between ambient gas and liquid fuel. A second important parameter is the aspect
ratio L/d0 of the nozzle hole, which is however not present in all correlations (practical values are
fairly constant, around 5; therefore this is often not systematically varied).
Dependence of cone angles on the kind of ambient gas has been investigated by Di Stasio and
Allocca [23]. They conclude that the cone angle is mainly governed by the density ratio, and to a
lesser extent by gas viscosity (however, their result shows that linearly increases with viscosity, in
contrast to what Eq. (19.7) suggests). Yet, since the composition of the ambient gas in diesel engines
is always very close to air, this eect is not too important for our present purposes.
In summary, use of the above correlations might provide a good estimate of cone angles, but their
absolute accuracy should not be overestimated.
364
19.2.3

Penetration
As briey discussed in the introduction (and in a great many papers, cf. [52, 79, 6, 23, 42, 21, 54, 47] to
name just a few), spray penetration can be characterized as follows. After leaving the nozzle, the spray
quickly atomizes and the resulting mixture of (very small, continuously evaporating) droplets and
entrained air protrudes the combustion chamber. The head of the spray continuously meets stagnant
air, and as a result develops a head vortex. A typical example is shown in Fig. 19.5, together with
the denitions of penetration length S and cone angle .
Behind the head vortex, spray edges are on average relatively straight, with the exception of
intermittently developing smaller side vortices. This region is commonly denoted as quasi-steady:
although it grows in length (following behind the head vortex), processes in that region are again
on average, i.e. apart from turbulent variations invariant with time during the steady phase of
fuel injection. The latter typically occurs for Common Rail injectors, which show an almost top-hat
injection prole, cf. [70, 21]. Since the fuel decelerates in the head vortex, the central velocity in the
quasi-steady region is higher than that of the spray tip [21].
Figure 19.5: Picture of penetrating fuel spray (EHPC data) showuing penetration, cone angle and liquid length
Most important for the diesel process is the speed of spray penetration. Two regimes can be
distinguished. Initially, the spray leaves the nozzle hole with a Bernouilli-like velocity; there has
not yet been signicant air entrainment. As a result, the velocity is initially constant and S (t) t.
At later stages, further from the nozzle, we have a constant cone angle, which means that there is a
constant rate of air entrainment (see section 19.4). As a result, the overall mass contained in the spray
in each slice is proportional to its position x. Due to conservation of momentum (total mass times
velocity = constant), we have xdx/dt = const., leading to x t1/2 . In other words, further away from
the nozzle, spray penetration goes with the square root of time due to momentum conservation.
In between these limiting cases, there is a transition from the linear (dominated by liquid) to the
square root (dominated by entrained gas) regime. This transition has been quantied dierently by
dierent authors, as will be discussed later. The experimentally observed transition time is usually on
the order of 0.5 ms. This is comparable with typical opening times of a diesel injector; consequently,
it is not fully clear whether the observed transition is due to the above mechanism or to nite injector
response, as Desantes et al. argue [22]. In the latter case, it is not fully clear why the initial S (t)
transient would be exactly linear in time.
In many publications, the spray tip penetration S (t) is tted as a power of time. Logically, in view
of the above, exponent values are usually found slightly above 0.5, cf. [79, 22, 43].

19.2.4
365
Evaporation
The above considerations on spray penetration hold for the fuel in general, independent of its phase.
That is, for non-vaporizing (liquid) sprays the same overall behavior is observed as for vaporizing
sprays (although the method of observation diers; liquid sprays can be observed using shadowgraphy,
whereas the vapor phase can only be visualized using Schlieren methods). The reason is that spray
penetration is purely determined by dynamics (masses and velocities of fuel and air).
In case of a vaporizing spray, the vapor phase shows the same tip penetration characteristics from
the previous section. The liquid phase initially travels together with the vapor phase [61], but after
some distance it stops; that is, no liquid is observed anymore. This distance remains quasi-steady
(again apart from turbulent uctuations) during the steady fuel injection phase, and is commonly
denoted as liquid length L [74, 79, 54].
The observation of such a quasi-steady liquid length is due to continuous air entrainment. At a
certain distance from the nozzle, there is a balance between the amount of heat entrained by the hot air,
and the amount of heat required for full vaporization. Quantication of this balance will be discussed
in section 19.4.
Much attention has been devoted to measurement and interpretation of the quasi-steady liquid
length in diesel fuel sprays, since it largely determines possible liquid fuel impingement on cylinder
walls (known as wall-wetting) [11]. However, looking into more detail, also a liquid width exists:
at a certain distance from the nozzle, liquid is only present up to a certain radial position as well.
This establishes itself in the liquid phase showing a smaller apparent cone angle than the vapor phase
(cf. [74, 28, 43], also visible in Fig. 19.5. In phenomenological models that include the radial spray
direction, this behavior is accounted for, see section 19.4.4. It is important if one aims to make detailed
predictions of the overall equivalence ratio and temperature eld in the vicinity of the spray.
Conceptually, the quasi-steady liquid penetration is ultimately determined by droplet evaporation.
However, there are many cases in which only the end stage of droplet evaporation is important, since
evaporation transients are typically much faster than the timescale of mixing in modern diesel sprays
(again related to the very small droplet sizes after breakup). This will be explained below.
Droplet evaporation has been discussed by many authors. Didactically, the book of Turns [78]
g
can be recommended (although he causes some confusion by using the gas mixture specic heat c p
g
where it should be fuel vapor specic heat c p f in Eqs. (19.10)-(19.11) below). The latter point is also
noticed in an excellent review paper by Sazhin [69], who discusses many aspects of droplet evaporation in great detail. In his conclusions, Sazhin states that the model by Abramzon and Sirignano [4]
is still widely used since it combines accuracy and computer eciency. A good example of its use is
the paper by Hohmann and Renz [36], which is of very good quality and resolves some small shortcomings and provides some updates to correlations used in Ref. [4]. Below, the equations governing
droplet evaporation are summarized, as combined from the references just discussed. Details of their
derivation are not provided, but can be obtained/discussed on request.
The mass ow m from an evaporating droplet of radius r s can be obtained from both mass diusion
and from energy conservation. Both routes result in dierent expressions, which are however coupled
via the temperature T s at the droplet surface [78]. These expressions are:
Sh0 (Re) 2
ln (1 + BM )
F(BM )
kg
Nu0 (Re) 2
m = 2 g r s 2 +
ln (1 + BT ),
F(BT )
cp f
m = 2g D f a r s 2 +
(19.10)
(19.11)
366
where g is the gas phase density, D f a is the binary diusion coecient of fuel in the ambient gas and
kg the thermal conductivity coecient of the vapor phase. The universal function F is dened by [4]
F(B) =
(1 + B)0.7
ln(1 + B).
B
(19.12)
The mass and heat transfer numbers BM and BT are given by

BM =
BT
Y f,s Y f,
1 Y f,s
g
c p f (T T s )
(19.13)
Hvap, f (T s ) + Qil /m
(19.14)
The saturated fuel vapor fraction Y f,s can be obtained from Eq. (19.20) below. Conditions far from
the droplet are denoted by subscript .
The internal droplet heating Qil is often neglected, in which case the heat ux to the droplet is just
balancing the mass ux from it times the heat of vaporization Hvap . For steady far eld conditions,
the droplet surface temperature T s is then steady (an eect that is for instance used in certain types of
household hygrometers). This so-called wet-bulb approximation is often reasonably accurate.
More advanced approaches to account for instationary droplet heating are discussed by Sazhin [69].
Amongst those, the eective conductivity model is recommended [3]. This model accounts for the increased heat transfer due to internal circulation within the droplet, by using an eectively higher liquid
heat conductivity coecient. Especially when combined with the parabolic model [24], the eect
of instationary droplet heating can be taken into account in an ecient and elegant way.
The Sherwood and Nusselt numbers above, Sh0 and Nu0 , are given by
Sh0 (Re) = Nu0 (Re) = 1 + (1 + ReSc)1/3 f (Re),
(19.15)
where
f (Re) = 1,
Re 1;
and
Re0.077 ,
1 < Re 400.
(19.16)
The Reynolds number Re is evaluated for the relative velocity between droplet ud and gas u , so
Re = 2r s |u ud |/g , where g is the gas phase dynamic viscosity. When the present evaporation
model is embedded in a spray model, the relative velocity is evaluated using the following equation
describing the droplet dynamics [36]:
dud
3C D
1 0.5
=
(u ud )2
dt
f
8r s f
(19.17)
This equation can be derived from the denition of the drag coecient C D , in combination with
elementary expressions for the mass and frontal surface area of the droplet. In Ref. [4] this equation
is a factor of 4 in error, which was repaired later in Ref. [3]. The term in brackets on the left hand side
of Eq. (19.17) accounts for the extra inertia that a droplet experiences in a gas [68]. Finally, the drag
coecient C D can be evaluated from
CD =
24
(1 + 0.15Re0.687 )(1 + BT )0.2 .
Re
(19.18)
Wherever uid properties of the gas mixture are needed in the above equations, these are to be
1
evaluated at temperature T = 2 T s + 1 T . When composition is needed, a value Y = 2 Y s + 3 Y is
3
3
3
recommended. This is known as Hubbards rule [37].
367
Obviously, quantitative results of the above model depend on the initial droplet size. For the latter,
sometimes correlations are used (see for example Ref. [40]); others use experimental data [46] or
compute it from detailed breakup models, cf. [35, 36, 38].
In spite of the complexity of the above equations, the results are easily summarized qualitatively.
Initially, when a droplet enters a hot, unsaturated environment, it will evaporate according to the socalled d2 law, i.e. the square of its diameter will decrease linearly with time. During this phase, the
droplet is in quasi-steady equilibrium with its surroundings: although its diameter is decreasing, its
temperature and evaporative mass ux are constant in time.
However, during this process, fuel vapor is continuously transferred to the droplet environment.
Since droplets are close within a spray, each of them only has limited space available and the result
is that its environment quickly saturates. Consequently, the rate of evaporation decreases and the
droplets size reaches an equilibrium value. This is qualitatively sketched in Fig. 19.6.
Figure 19.6: Droplet-in-a-box, evaporating down to the saturation limit determined by mixing
In conventional diesel spray modeling, the above process is usually modeled in full detail, which
gives rise to numerical issues (as discussed in the next section). However, as it appears, conditions in
modern diesel sprays are often such, that the time scale of droplet evaporation is much smaller than
that of mixing [76, 52, 16]. In Fig. 19.6, this implies that the droplet diameter very quickly falls
down to the equilibrium level indicated by the dashed lines. Further evaporation then only occurs
if that equilibrium level shifts it only does so due to additional air entrainment. For this reason,
this type of evaporation is denoted as mixing-limited vaporization, in which local transport processes
around the droplet are unimportant, in the sense that they are fast enough in order not to be limiting.
In the latter case, only the description of the limiting case is needed; this situation corresponds
to thermodynamic equilibrium between the droplet and its surroundings. It can be demonstrated that
this equilibrium state only depends on the properties and thermodynamic conditions of fuel and air,
and on the masses available of both (i.e. it does not depend on details of the droplet distribution).
This is illustrated in Fig. 19.7, which shows equilibrium vaporization in a system at constant pressure
pa (didactically corresponding to the in-cylinder ambient pressure), starting from masses m f and ma ,
each with their own initial temperature.
When ma in Fig. 19.7 increases, there will be a certain ratio ma /m f for which the ambient air
carries just enough energy to evaporate all the (initially liquid) fuel. For that situation, mevap = m f 0 .
368
Figure 19.7: Illustration of begin and end states of mixing-limited vaporization in a pressurized container
The end state is then, in view of the constant pressure, characterized by an enthalpy balance in which
the enthalpy gain of the fuel is equal to the enthalpy loss of the ambient gas:
g
m f [h f (T s ) hlf (T f )] = ma [ha (T a ) ha (T s )].
(19.19)
At the same time, the composition of the resulting mixture is characterized by saturation of the fuel
vapor at temperature T s (and, strictly speaking, at pressure pa although the latter eect is generally
small [44]):
M f p sat (T s )
mf
f
Y fsat (T s ).
(19.20)
=
ma + m f
Ma [pa p sat (T s )] + M f p sat (T s )
f
f
The preceding two equations form a coupled system for the saturation temperature T s and saturated
fuel mass fraction Y fsat . This will return in dierent forms in the mixing-limited phenomenological
models to be discussed later. Importantly, this system of equations does not contain any information
on the spray geometry; it is purely a thermodynamic equilibrium. Spray details come in via the local
mixing rate ma /m f , which is, as argued before, only governed by dynamic processes (fuel mass and
momentum conservation).
Before turning to a more quantitative description of fuel penetration and vaporization, we will
now briey discuss combustion in sprays, in order to complete our description of dierent phenomena
involved in a combusting diesel spray.
19.2.5
Combustion
Only a few decades ago, the picture of combustion in diesel sprays was dramatically dierent from
current understanding. As summarized by Dec [14], there was considerable debate whether fuel
droplets would burn individually (with a small ame around each of them) or in groups. For brevitys
sake this discussion will not be repeated here.
Based on a number of laser diagnostic studies, available roughly since the 1980s, the current
understanding is that the liquid droplets have already fully evaporated in regions where ames exist,
as pointed out earlier in this chapter. This holds both for the rich premixed ame upstream of the
spray and for the diusion ame surrounding it. For this reason, the main thing to understand is how
atomization and droplet evaporation aect the resulting (, T ) temperature and equivalence ratio
elds in the spray. Accordingly, recent spray models often only deal with the liquid phase to derive
eective source terms of fuel vapor, momentum and energy. This will be more extensively discussed
in section 19.3.
369
The most important phenomena in spray combustion are auto-ignition, ame stabilization (and the
associated lift-o length) and formation of emissions (mainly NO x and soot/PM, particulate matter).
Auto-ignition takes place when the rst part of the injected fuel has mixed with sucient air to reach
the right (, T ) conditions. This process has been dealt with in the previous chapter and will not be
addressed in any more detail here (although its importance can hardly be overestimated!).
During the quasi-stationary phase of spray combustion, the rich premixed ame, initially resulting
from auto-ignition, is stabilized at a certain distance H from the nozzle (again quasi-stationary, i.e.
high-frequently varying due to turbulence). This distance, known as the ame lift-o length (FLoL)
has been extensively measured and correlated as a function of temperature, density, nozzle diameter
and injection pressure by the Sandia group [72]. From their results, Pickett et al. derive the following
correlation for normal diesel fuel [64]:
3.74 0.85 0.34 1 1
H = 7.04 108 T a
a d0 U0 Z st ,
(19.21)
where H is in [mm], d0 in m and other quantities have their normal SI units. In additional papers,
the eects of ambient oxygen concentration [73] and fuel oxygen [49] on H were investigated.
For measurements of H, chemiluminescence of OH radicals is used, a method that is critically
evaluated in Ref. [71]. It must be noted that observations of lift-o lengths are sometimes reported in
terms of natural soot luminosity of the ame, also known as the soot lift-o length (SLoL). Since
soot particles need nite time to form and grow before they signicantly radiate, SLoL is always larger
than FLoL and thus provides an overestimate of the real ame lift-o length based on OH as a ame
front marker.
Higgins and Siebers in Ref. [71] also compare their values of H to previously measured values
of the liquid length L. They conclude that either of the two can be larger than the other, depending
on prevailing conditions, see Fig. 19.8 for an illustration. Increasing ambient density and temperature
tend to make H smaller than L. On the other hand, higher injection pressure, smaller nozzle holes and
increasing EGR levels have the reverse eect. The latter three are the main trends in modern diesel
engines hence, it seems that the trend is towards H > L.
Figure 19.8: Flame lift-o H versus liquid length L. Left: H < L; T a = 1100 K, a = 23 kg/m3 , pi = 40 MPa
and d0 = 250 m. Right: H > L; T a = 1000 K, a = 20 kg/m3 , pi = 200 MPa and d0 = 100 m. Adapted from
Ref. [72].
The above has important implications for the complexity of spray models; when H < L, it is
essential to include the radial coordinate of the spray in modeling, as is for instance done in the
370
Valencia spray model, see section 19.4.4. In that case, the mutual inuence between evaporation and
combustion is obviously large, since the ame surrounds the liquid core.
Considering the mechanism of ame stabilization at x = H, there is still considerable debate. An
explanation originally suggested by Kalghatgi [41] is that, at the ame base, the upstream turbulent
burning velocity just balances the downstream convective spray velocity. This hypothesis was used in
Ref. [72] to correlate the results, based on a theoretical relationship by Peters [60] derived from this
hypothesis. However, it appears not to explain all observated trends. Specically, it does not explain
the observed large dierences in H between fuels (burning velocities of most fuels are very similar!).
An alternative explanation has been suggested by Pickett et al. [64], who convincingly establish
a connection between the observed lift-o length and the ignition delay of the fuel. It comes down
to assuming that H establishes at a location where newly incoming fuel has had enough time to
auto-ignite. This is conrmed by the observation of intermittent ignition pockets slightly upstream
of H. The auto-ignition hypothesis seems to be conrmed by recent measurements in which forced
ignition is applied [61], although the latter authors ascribe their observations to re-entrainment of hot
combustion products, rather than auto-ignition. This mechanism is also mentioned in Ref. [64].
Clearly, the discussion on the stabilization mechanism of H has not yet been nished. A recent
suggestion is that, at the prevailing high temperatures in diesel engines, ame propagation chemistry
and ignition chemistry become more similar, so there would be less of a contradiction than suggested
above (Kalghatgi, private communication, 2010).
The equivalence ratio of the rich premixed ame near x = H is generally between 2 and 4. The
resulting products of incomplete combustion (partially oxidized fuel, CO and soot precursors) travel
further downstream, in the hot core region of the spray, where ideal conditions for soot formation and
growth exist [63]. The resulting soot and remaining products are subsequently oxidized for the largest
part in the diusion ame surrounding the spray although surrounding may not be the right word;
due to turbulence, the diusion ame front is heavily distorted and wrinkled.
The diusion ame is also the main location of NO formation (which in the exhaust pipe partly
transforms into NO2 ). The amount of NO is mainly governed by the adiabatic ame temperature T ad .
This in turn is mostly governed by in-cylinder conditions (intake temperature and EGR percentage),
and is less aected by other fuel parameters or spray details.
From the above description it will be clear that the amount of exhaust-soot is very much aected
by details of spray combustion. It has been shown that the amount of soot formed depends strongly on
the equivalence ratio of the rich premixed ame (H). Whenever (H) is smaller than 2, no signicant
exhaust-out soot is observed [73, 49]. This might be tentatively explained by the hypothesis that, on
average, each C atom oxidizes to CO when = 2, thereby preventing it from going into soot precursor
molecules.
If (H) > 2, a strong correlation of its value with the amount of soot in the spray is observed [73].
This will be quantied more in section 19.4.2. However, the amount of exhaust-out soot in that case
was recently found to be limited by oxidation rates rather than (H) [5].
Summarizing, the ame lift-o length H is an important parameter for soot emissions, since it
determines the amount of air entrainment upstream of the premixed ame, at the foot of the burning
diesel spray. This will be further quantied in section 19.4.2.
19.3
371
Classication of spray models
In view of the enormous complexity of the spray process in engines, it will come as no surprise that
many attempts have been made to provide simplied models of (parts of) the process. These models
fall apart into dierent categories.
19.3.1
Heuristic models; correlations
The rst and oldest category are empirical correlations, i.e. equations that are capable of reproducing
experimental observations of sprays. Such models make no attempt to explain the observations; they
merely describe them, although the chosen mathematical form is preferably based on some physical
insights. Therefore, these models are named heuristic models, or just correlations. Examples were
already shown in the previous section. Eqs. (19.6)-(19.9) were derived purely based on observations,
without any attempt to explain the magnitude of the powers. The same holds for the correlation of
Sauter mean diameters of droplets, Eq. (19.5).
Since correlations are often not more than curve ts (polynomial, exponential or power laws),
they can only be used in the range where they have been tted. As a matter of fact, this rule is often
overlooked in practice. There are many examples of ts being abused outside their tted range,
and often this is hard to notice. CFD spray models often contain submodels in the form of empirical
correlations, without clearly pointing out their limitations in the manual. However, this remark is not
exclusive to spray modeling; it holds wherever descriptive models are used to make predictions.
As a matter of fact, a myriad of correlations has been used to describe macroscopic observations
of sprays. As an illustrative example, one may look at Ref. [29]. By then, in 1985, already 21 dierent
correlations for spray penetration were listed, 4 for SMD, 9 for cone angle and 9 for ignition delay. The
needed parameters are not very well specied in that paper. Moreover, it must be noted that Ref. [29]
is fairly old. More recent books, e.g. those of Stiesch [77] and Baumgarten [8] are good sources of
more recent correlations. Still, in view of the fact they should always be used with care, empirical
correlations are generally unattractive for studying spray combustion over broader parameter ranges.
Especially when models are extrapolated to conditions that have not yet been tested one of the main
purposes of modeling! one should be careful.
19.3.2
Phenomenological models
As a second class, phenomenological models are model descriptions based on some understanding of
the underlying physics of the problem. However it is recognized that the physics can not be covered
in full detail. Consequently, simplifying assumptions are made to make the model mathematically
tractable.
As an example, for a spray, one can look at integral conservation laws to draw conclusions about
the scaling of penetration with density or other parameters. The resulting models are often analytical
or semi-analytical; that is, they can be expressed in a limited number of equations that can be solved
using a pocket calculator or, at most, a simple computer (as opposed to CFD computations).
Phenomenological models are thus appealing, since they combine physical insight with computational simplicity. They also allow some (cautious) extrapolation, which enables parameter studies
over wider parameter ranges (model-based extrapolation is far safer than using e.g. polynomials!).
Last but not least, phenomenological models of evaporating fuel sprays can be used to provide source
terms for the gas phase in CFD models, thereby avoiding the notorious grid dependency problems
associated with conventional spray CFD models, which will now be discussed.
372
19.3.3

Computational Fluid Dynamics
A third and nal category of spray models are those based on Computational Fluid Dynamics (CFD).
In CFD modeling, the transport equations of mass, momentum and energy (and, in combustion systems, also species) are solved for every single cell in the computational grid. Therefore, far less
simplication is needed. The associated drawback is, of course, that CFD models are by far the most
demanding in terms of complexity and computation time.
Particularly for sprays, there are other complications at hand. As argued in, for example, Refs. [80,
39, 38, 77, 8], it is widely observed that the results of spray CFD modeling principally suer from
grid dependency. This holds both for the grid cell size and for the structure of the grid. The reasons
are twofold: lack of spatial resolution in the vicinity of the nozzle, and lack of statistical convergence
in the treatment of the liquid phase [8].
In order to do meaningful computations, the grid size should preferably be smaller than the typical
nozzle dimension. Abraham [1] has shown that at least 4 grid cells should t within the nozzle hole.
The latter being of the order of 0.1 mm, this clearly forms a limitation for computations of practical
sprays in realistic combustion chambers (with a size of the order of 0.1 m) a dierence of 3 orders
of magnitude in length scale, which is 9 orders of magnitude in volume!
Moreover, it is very dicult to compute the complex two-phase phenomena inside a spray. Generally, the gas phase is computed in an Eulerian way, making use of a xed grid. The liquid phase is
kept track of in a separate (Lagrangian) model. Such Euler-Langrangian models are still the default
approach [8]. Of course, mass, momentum and energy are exchanged between the two, which requires
detailed knowledge both on the droplet population (size, temperature, velocity etc.) and the exchange
mechanisms.
Moreover, the Lagrangian approach requires that the void fraction of each Eulerian cell is close to
one, i.e. the fraction of liquid in each cell must be small (not larger than about 10%). This requirement
is clearly not met for small grid cells in the immediate vicinity of the nozzle, which would be mostly
lled with liquid. To get around it, grid cells are used that are suciently large, so as to reduce the
amount of liquid in each cell in the vicinity of the nozzle. This is clearly in contradiction with the
requirement of suciently small grid cells.
The result of the above is that the submodels used for spray vaporization become dependent of
the grid size. To overcome this, it is common practice to tune parameters of submodels in order to
give meaningful overall results. However, the resulting tuned parameters often do not have physically
meaningful values any more. As Stiesch formulates it [77]: submodels are trimmed to an unphysical
behavior in order to overcome the deciencies caused by inadequate grids.
In recent years, a new approach has emanated which is able to overcome the grid dependency
issue. In this so-called Euler-Euler approach, the liquid phase is no longer explicitly followed.
Instead, both vapor and liquid phases are present within the same computational grid. This requires
modeling of the vaporization process, in order to distribute the fuel in each cell over both phases.
This can be done either in a droplet-limited or mixing-limited fashion.
Examples are the work of Abraham and Magi [2], Wan and Peters [81], Iyer et al. [39, 38] (who
also provide a rst exploration of the limits between both regimes) and of Versaevel, Motte and
Wieser [80]. The latter authors present a mixing-limited model of spray vaporization that allows
the use of an all-Eulerian model, with source terms based on a phenomenological model of the fuel
spray. Their model will be covered in more detail in the section to come. In that section we will only
discuss the submodels themselves; for the way of implementing them into CFD models, refer to the
corresponding papers.
19.4
373
Phenomenological spray models
In this section the most important (semi-)phenomenological spray models will be discussed in more
detail. The treatment will necessarily be concise avoiding derivations and lengthy explanation, only
mentioning the main points and equations. For more details the reader is referred to the original
papers.
19.4.1
The Hiroyasu model
One of the rst successful models of diesel spray combustion is the package model by Hiroyasu
and coworkers. Introduced in 1976 [33], it was later extended and improved several times [34, 82].
In this model, a diesel spray is divided into a number of packages, each containing a xed amount of
fuel mass. The amount of air in each package changes due to air entrainment. A schematic picture is
provided in Fig. 19.9.
Figure 19.9: Schematic picture of Hiroyasus package model of spray combustion
Penetration
Penetration of the spray tip S (t) is described by
S
= 0.39
= 2.95
2p
t,
l
p
a
0 < t < tb
(19.22)
t tb
(19.23)
1/4
d0 t,
where tb , denoted as the breakup time, indicating the transition between the regimes dominated by
liquid penetration and by air entrainment, is given by
tb = 28.65
l d0
.
a p
(19.24)
The constants in the above equations can be generalized to account for a variable nozzle discharge
coecient Cd , as demonstrated by Jung and Assanis [40]. This results in replacing 0.39 with Cd in
2
Eq. (19.22) and replacing 28.65 with 4.35/Cd in Eq. (19.24). These authors also provide a very useful
and detailed explanation of their implementation.
374
The above equations hold for the central packages of the spray. Corrections are provided for
packages at the periphery, which move more slowly, and for the eect of swirl on penetration. These
corrections can be found in the original papers and will not be reproduced here.
Vaporization
Air entrainment into each package is derived from momentum conservation: m f U0 = (ma + m f )U,
resulting in
U0
ma = m f
1 .
(19.25)
dS /dt
Hence, the air-fuel ratio in each package follows from the penetration speed of each package and the
initial velocity U0 at the nozzle. Again, corrections are made to this equation (by introducing simple
correction constants) to account for the eects of ignition and wall impingement. For convenience,
the dierential air entrainment rate can be derived from Eq. (19.25), giving
ma =
m f U 0 d2 S
.
(dS /dt)2 dt2
(19.26)
The fuel mass in each package can be combined with the Sauter Mean Diameter of the droplets
(which follows from separate measurements, cf. [32]) to provide the number of droplets in each
package. Given the air-fuel ratio of each package, droplet evaporation is then computed from the
equations provided in section 19.2.4.
Liquid length is not explicitly addressed in the Hiroyasu model, although it can be obtained from
following the packages on the central line and establishing the point in time (and, via S (t), position)
where the droplets have fully evaporated.
Historically, package models have always been combined with droplet-limited vaporization. However, there is no reason why mixing-limited vaporization could not be used in a package model. This
would greatly simplify the model, since the amount of fuel evaporation for each package over time
would in that case simply be obtained from a combination or air entrainment, described by Eq. (19.25),
and thermodynamic equilibrium obtained from Eqs. (19.19) and (19.20), both for mv instead of m f as
f
long as not all fuel has evaporated.
Combustion
The equivalence ratio in each package is obtained from the vaporized fuel mass, resulting from
droplet evaporation, and the entrained air. The value of each package is used to predict soot and
NO x emissions. For the latter, simple correlations (available by that time) were used, which are not
reproduced here for reasons of brevity.
These correlations also involve temperature, which is obtained for each package taking into account evaporation and mixing (air entrainment). The ambient temperature T a is kept track of, taking
into account the engines compression and expansion process, and heat loss to the cylinder wall. The
latter is also described using empirical correlations. The same holds for the ignition delay ID , which
identies the time after which combustion sets in (i.e., auto-ignition is not modeled explicitly). Again,
a similar approach is followed in Ref. [40].

19.4.2
375
The Sandia model
Penetration
Sandia researchers have developed a spray model that is able to correlate many of their measurements. A large database is available online, see https://share.sandia.gov/ecn/sprayCombustion.php.
The model started with the work of Naber and Siebers on spray penetration [52], and was later extended to include liquid length [75] and equivalence ratio at lift-o [72]. Later it was used in an
eort to correlate ignition delays to residence times for measured lift-o length [64]. Below the main
equations are reproduced. Not all symbols are explained, since most of them have occurred earlier.
Figure 19.10: Schematic picture of the Sandia spray model with radially uniform proles
In the derivation of the Sandia model, an idealized geometry is used, represented in Fig. 19.10.
The idealized spray is characterized by constant composition and velocity over each cross-section at
position x. The spray tip velocity is, within this model, assumed to equal the cross-sectional average
velocity v(x) = dx/dt. Starting from fuel mass and momentum conservation, Naber and Siebers [52]
derive an equation for the dimensionless spray tip velocity:
dx
2
=
.
dt
1 + 16 x2 + 1
(19.27)
This equation can be analytically integrated to give
t=
S S
+
2 4
1 + 16S 2 +
ln 4S +
16
1 + 16S 2 .
Dimensionless position x (for spray tip: S ) and time t have been introduced, according to
f
x
Ca d0
x =
, x+ =
x+
a a tan(/2)
f
t
1
Ca d0
+
t = +, t =
t
U0 a a tan(/2)
(19.28)
(19.29)
(19.30)
In these equations, spray cone angle correlations derived from the same set of measurements are used.
For vaporizing conditions, these are given by Eq. (19.9). The observed values of are multiplied with
a constant value a to compensate for the eect of the idealized spray geometry.
376
The value of a in Eq. (19.28) was rst empirically tted to 0.66, and later adjusted to 0.75 [73].
The latter value is in accordance with theoretical considerations, in which radial proles of fuel mass
and velocity are also considered. This is demonstrated in the appendix of Ref. [52]. (However, their
Eq. (C23) is in error: both A(x) and A(x) should be omitted from it, as conrmed by Dennis Siebers
in private communication). More importantly, it thus appears that the constant a in a way accounts
for the more realistic radial proles in the spray. In section 19.4.4 we will see that the Valencia model
explicitly takes these into account.
It can shown that Eq. (19.28) reduces to the well-known limits S t for small values of t,
be
and S t for large t. The transition between these regimes occurs for t 1. Since the equation
rather than S , it would often be easier to work with the inverse relation. This can
is explicit in t
however not be obtained analytically from Eq. (19.28); as an alternative, it is shown in Ref. [52] that
the relation
1/n

S = t 1 + tn/2
(19.31)
gives results very close to those of Eq. (19.28) for n = 2.2.
Vaporization
As in the Hiroyasu model, vaporization modeling must take into account air entrainment. As a spin-o
from the penetration model, the fuel-air mass ow ratio at (dimensionless) position x can be shown to
obey
m f ( x)

2
=
.
(19.32)
ma ( x)

1 + 16 x2 1
At the same time, again from thermodynamic considerations, the fuel-air ratio can be obtained from
an enthalpy balance at the position where all liquid has just disappeared (i.e. the liquid length L).
From that enthalpy balance, Siebers shows that
B
m f (L) ha (T a , pa ) ha (T s , pa p s )
p s (T s )M f
Za
=
=
.
ma (L)
h f (T s ) h f (T f , pa )
[pa p s (T s )]Ma Z f
(19.33)
From the nal equality, the saturation temperature T s can be obtained; the value of B follows from the
same computation.1
Knowing B and inserting it into Eq. (19.32) for x = L = L/x+ , it follows that

L = 0.25
In dimensional form,
b
L=
a
2
+1
B
f
Ca d
a tan(/2)
1.
2
+1
B
(19.34)
1,
(19.35)
where the 0.25 has been substituted by constant b. The latter is tuned by Siebers in Ref. [75] to a value
of 0.41. A more elaborate discussion on the model constants a and b is given in Ref. [45]. It turns
out that, within the cross-sectionally averaged models, values obtained for b are fairly constant [80].
However, the need of a tunable constant is a serious shortcoming of the model. In the Valencia spray
model, using radial distributions of fuel vapor fraction and velocity, it appears that such constants are
not needed to achieve satisfactory agreement with measured liquid lenghts [16].
1
The ratio of compressibility factors appearing at the end of Eq. (19.33) is an actually wrong attempt to account for
real gas eects. In Ref. [45] it is shown how this can be done more consistently. The same holds for the pressure dependency
of the enthalpy terms in this equation.
377
Combustion
From Eq. (19.32), the fuel-air ratio is in principle known at every position x. However, this ratio
pertains to the total fuel content, not only the vapor. Downstream of the liquid length, fuel only exists
in vapor phase, and in that case the fuel-air ratio of Eq. (19.32) equals that of the vapor phase. In
section 19.2 we have seen that, for modern diesel conditions, the ame lift-o H is often larger than
L. In that case, we can use the fuel-air ratio at H to estimate the local equivalence ratio:
(H) =
2AF st
1 + 16(H/x+ )2 1
(19.36)
As discussed in section 19.2, the equivalence ratio at H has important consequences for soot formation. Whenever (H) < 2, no signicant soot formation is found. This was hypothetically attributed
to every C-atom being bound to 1 oxygen atom.
This observation was later extended [48] to include oxygenated fuels, introducing the total atomic
oxygen ratio . The latter is a combination of fuel oxygen and the conventional equivalence ratio,
and accounts for the total amount of available oxygen with respect to stoichiometric:
(H) = f +
1 f
.
(H)
(19.37)
In this equation, is the equivalence ratio resulting from air entrainment. f = z/(2x+y/2) represents
the amount of oxygen available in the fuel, over the amount that is stoichiometrically needed for an
oxygenated fuel characterized by C x Hy Oz . It appears that all fuels tend to zero soot when becomes
larger than 50%. However, the initial soot level (and thus the slope) diers per fuel [62, 48], as
depicted in Fig. 19.11, adapted from Ref. [48].
Figure 19.11: Observed soot versus total atomic oxygen ratio at ame lift-o (H), adapted from Ref. [48]
The above method of predicting local equivalence ratios was rst used explicitly for combustion
modeling by Flynn et al. [25]. They divided the combusting spray into zones of varying temperature
and equivalence ratio, and applied detailed chemical kinetics for each zone. Although this approach
was still rather crude, it did explain some of the observed trends in soot formation. Their Figs. 10 and
16 (not reproduced here) nicely summarize their main ndings.
In recent years, the Sandia model has been used by many authors. As an example, the work of
Bayer and Foster is mentioned [9], who combine the above model with simple emission formation
models (very similar to those employed earlier by Hiroyasu and coworkers [34, 82]) to compare with
experimental engine data.
378
19.4.3

The Versaevel model
The Sandia liquid length scaling law, summarized in the previous section, only provides information
on the liquid length position. No information is available on the evaporation (hence fuel content of
the mixture) upstream of L. This point has been addressed by Versaevel, Motte and Wieser [80].
Introducing the void fraction m, they account for vaporization upstream of L. This is done assuming mixing-limited vaporization at any position x upstream of L, not only for L itself. The model is
embedded in the same context as the Sandia model, being an idealized spray with radially uniform
fuel fraction and velocity proles (as depicted in Fig. 19.10).
Introducing the mass ow ratio (equivalent to 1/B in the Sandia model), Versaevel et al. derive
from fuel mass conservation:
m(1 Y f g )g
ma
=
=
,
(19.38)
m f 0 f (1 m) + mY f g g
where Y f g is the mass fraction of fuel in the gaseous phase. The parameter g represents the density
of the gaseous mixture, containing fuel vapor and ambient gas.
From total momentum conservation, a similar expression follows:
2 =
Am2 (1 Y f g )2 2
g
A0 f 0 [(1 m) f + mg ]
(19.39)
The surface area ratio A/A0 is fully determined by geometry. For each position x, its value equals
A/A0 = 1 + (2x/de )a tan(/2), with de = Ca d0 .

The law of energy conservation in this framework reads
m f 0 h f l (T f ) + ma ha (T a ) = f (1 m)Avh f l (T ) + Y f g g mAvh f g (T ) + ma ha (T ),
(19.40)
which can be rewritten in terms of specic heats as

Ta
c pa (s)ds =
Y f g
1 Lv (T ) + Lv (T f ) +
1 Yfg
c p f g (s)ds.
(19.41)
Tf
Herein c pa is the specic heat of the entrained ambient gas and c p f that of the fuel vapor. Lv represents
the latent heat of vaporization. T denotes the temperature of the gas-vapor mixture at axial position x.
The three basic conservation laws (19.38), (19.39) and (19.41) contain the unknowns , m, g ,
Y f g and T . Upstream of the liquid length, the assumption of thermodynamic equilibrium couples the
mass fraction of fuel vapor to temperature. The fuel partial pressure equals the fuel saturated vapor
pressure: X f g = p s /pa , where X f g is the saturated fuel molar fraction. Converting it to a mass fraction
notice the similarity with Eq. (19.20) gives
1
Yfg =
.
(19.42)
pa
M
1 Ma + 1
p s (T )
f
Assuming uniform pressure, g can be linked to T and Y f g via the ideal gas law, yielding
g = a
1
Ta
.
Ma
T Y f g + [1 Y f g ]
Mf
(19.43)
Using the three non-linear conservation equations (19.38), (19.39) and (19.41), and aided by Eqs.
(19.42) and (19.43), the unknowns (T , and m) can be found using standard numerical methods.
Downstream of the liquid length, te void fraction m is unity and Y f g becomes smaller than its saturated
value, due to continuous air entrainment. As a result, Eq. (19.20) is no longer valid, and Y f g becomes
an unknown in Eqs. (19.38), (19.39) and (19.41), with m = 1 in that case.
379
Penetration
An excellent description of the Versaevel model and its implementation is given by Bekdemir [10].
He also indicates how, from the obtained solutions sketched above, the velocity v(x) can be obtained.
This is not trivial, since the model in principle provides quasi-steady solutions at each position x.
As in the Sandia model, the spray tip velocity is assumed to equal v(x). Hence, S (t) follows from
integration of v(x), keeping track of the tip position:
S (t + dt) = S (t) + v[S (t)]dt.
(19.44)
As also indicated by Bekdemir, the results of the model strongly depend on the chosen spray angle that
needs to be put in. Versaevel et al. use the same spray angle correlation Eq. (19.9) as Siebers [75],
eectively multiplied with the same constant a to account for the eect of the (unrealistic) idealized
spray with cross-sectionally uniform variables.
Vaporization
The liquid length in the Versaevel model follows from the model results as the position where m = 1
rst occurs. In principle, vapor and liquid penetration are computed simultaneously in this model.
However, in order to achieve correspondence with liquid length data from Verhoeven et al. [79], the
product of model constants ac needs to be adjusted from 0.23 to 0.104 (c = 0.27, see Eq.(19.9)) for
liquid length computations. The authors argue that this is in accordance with the Sandia model, which
also needs a specic correction factor b to achieve correspondence with experimental data.
Consequently, the model is run twice, with dierent values of ac. The smaller value of ac provides
the steady liquid length L; the larger value provides transient penetration. Since vapor and liquid are
known to penetrate together during the transient phase, the total liquid penetration xl is evaluated as
xl (t + dt) = min(S (t + dt), L).
(19.45)
Although the resulting constants of the Sandia and Versaevel models are indeed equivalent [80,
45], it appears that liquid length results are not the same [10]. Luijten and Kurvers [45] provide an
in-depth analysis of the dierences between the models. In doing so, they derive a closed expression
for the liquid length in the Versaevel model:
1
L=
4a
f0
Ca d
a (L) tan(/2)
2
+1
B
1
Ca d
1
.
2a tan(/2)
(19.46)
This equation bears great similarity to Eq. (19.35). Apart from the (small) last term, which corrects
for the eect of nite nozzle size, the main dierence is in the evaluation of a . In Eq. (19.46) it is
evaluated at L, with temperature T s instead of T a . Physically, this dierence accounts for the eect
of upstream vaporization on penetration. Although at x = L a considerable amount of fuel vapor is
present, lowering the partial density of air, the temperature eect is dominant. The result is a higher
a value, and hence smaller liquid lengths.
Another important conclusion from Ref. [45] is that the thermodynamics in both models are fully
equivalent, i.e. the same values of B resp. 1/ and T s are found. This conrms the earlier conjecture
that the saturation thermodynamics can be decoupled from the process of spray penetration, which is
purely governed by dynamics. Still, the above argument suggests that there can be a one-way coupling
from thermodynamics to spray penetration. This needs to be investigated further.
380
Combustion
The Versaevel model was introduced to avoid the notorious grid dependency problems with EulerLagrangian spray models. The model introduces source terms of fuel mass, momentum and energy
into an Eulerian CFD model of the gas phase. This is done only for the region up to the liquid length
L; after that region, the CFD model takes over (the source terms become zero anyway for x > L).
Bekdemir [10] has investigated this approach more closely and writes down the resulting source
terms explicitly. The time derivative of liquid fuel mass is
m f l (x) = f l [1 m(x)]A(x)v(x),
(19.47)
where all quantities can be obtained from the spray model. Division by a volume element A(x)d(x)
and changing the sign gives the source term of fuel vapor mass per unit of time and volume:
mfv =
1 dm f l (x)
.
A(x) dx
(19.48)
Similarly, the momentum source term is obtained from multiplying Eq. (19.47) by v(x):
1 d[m f l (x)v(x)]
.
Mfv =
A(x)
dx
(19.49)
The energy source term is slightly more complex, since energy is transferred in both ways. A negative
enthalpy source term represents liquid heating, whereas a positive one accounts for added fuel vapor:
1 dE f l
Eg (x) =
+ m f v E f g (x),
A(x) dx
(19.50)
where E f i = h f i pa / f i (x) + v(x)2 /2 with i = f, g for both phases.

Using this approach, Bekdemir [10] was able to couple the Versaevel model to a commercial CFD
code. Combustion can then be modeled using any method of choice; this is fully decoupled from the
spray model, since the latter only aects gas phase composition and temperature. For this reason,
combustion modeling is not further addressed here. An example result is shown in Fig. 19.12.
Figure 19.12: Temperature contours of combusting n-heptane spray, using two dierent combustion models
(eddy-dissipation model and FGM). Taken from Ref. [10]. Notice the dierent ame-lift o positions

19.4.4
381
The Valencia model
In this section the spray model, developed over the last decade at the Universidad Polit cnica de
e
Valencia, is discussed. Like the models discussed above, it is semi-phenomenological in nature; i.e.
not solving the full 3D equations, but simplifying the physics to an extent that overall relations can be
deduced for e.g. spray tip penetration and liquid length. This enables to establish relations between
these phenomena and, for instance, density, temperature, injection pressure and nozzle characteristics.
Concerning the latter, the same group has published a lot of work on the determination of nozzle
characteristics and their eect on spray development. An important general conclusion of that work
is that nozzle details, via the spray cone angle, play a very essential role in prediction of macroscopic
characteristics. Therefore, generalization of modeling results is very hard without measuring nozzle
discharge characteristics Cd , Ca and Cv , cf. Ref. [6].
To connect to the models discussed earlier, it helps to realize the following. Both the Sandia and
Versaevel models are able to compute certain properties for every x-position in the spray. It might
seem that this is only possible due to the assumption that these properties are uniform over each
cross-section; however, if the functional form of the radial prole is prescribed, it is possible to obtain
radial proles from integral conservation laws.
More specically, Gaussian proles are used; besides allowing easy integration, a physical reason
for this lies in the Central Limit Theorem. This states that a combination (convulation) of many
arbitrary statistical distributions will always give an overall Gaussian prole. Also experimentally,
Gaussian proles are always observed for velocity and fuel mass distribution.
Penetration
The Valencia spray penetration model, rst published in Ref. [21], starts from conservation of momentum for every position x:
R
M(x) = M0 =
2(x, r)rU 2 (x, r)dr.
(19.51)
The momentum ow M0 at the nozzle is an input to the model (and can be measured as discussed
before). The velocity eld is prescribed as
U(x, r) = Uaxis (x) exp
r
R
(19.52)
The shape factor , related to the (measured) spray angle via the outer radius R = x tan(/2), describes
the radial width of the velocity prole. Taking U(x, R) = 0.01Uaxis , its value is 4.6. In principle, this
self-similar velocity prole only holds in the conical (quasi-steady) part of the spray, constituting
about 70-80% of the spray length.
The density in Eq. (19.51) complicates the integration. Therefore, in Ref. [21] a transformation
is done to an iso-density spray, having the same exit velocity and momentum ux, but with a (larger)
eective diameter deq = d0 f /a . In that case integration is straightforward and results in
2
M0 = a Uaxis (x)[1 exp(2)]x2 tan2 (/2).

2
(19.53)
In view of the large value of , the factor in square brackets can be omitted. The resulting axial
velocity prole is
1/2
M0
.
(19.54)
Uaxis = 1/2
a (/2)1/2 x tan(/2)
382
Based on experimental as well as theoretical evidence, it is argued that the spray tip moves with a
velocity equal to one-half of the expected axial velocity at x = S , in other words dS /dt = 0.5Uaxis (S ).
Since Uaxis (S ) 1/S , this can be integrated, resulting in
S =
1/4
1/4 a
M0 1/4 t1/2 tan1/2 (/2).
(19.55)
From this result it appears that only the square root limit of S (t) is described. Although the authors
recognize that there is a (short) period where penetration is linear with time, they make no attempt to
model that; it is ascribed to the opening behavior of the nozzle, rather than a dierent ow regime [22].
In a later paper [19], the iso-density assumption is abandoned. Instead, a Gaussian prole is posed
for the (total gas plus liquid) vapor fraction as well:
Y f (x, r) = Y f,axis (x) exp Sc
r
R
(19.56)
The density (x, r) straightforwardly follows from this prole. Yet, this results in considerable mathematical complexity. However, if one assumes Sc = 1 and f
a (the resulting errors are quantied
in Ref. [19]), the resulting expression is not too complex:
Uaxis
M0 = x2 a tan2 (/2)U0 Uaxis U0 ln 1
U0
(19.57)
As before, Uaxis can be solved from this equation, in this case by iteration. For Sc = 1 it can also be
shown [21] that Y f,axis = Uaxis /U0 , in other words: the proles of dimensionless velocity and mass
fraction are identical. Consequently, having obtained Uaxis as above, both U(x, r) and Y f (x, r) are
known.
Vaporization
The distributions of velocity and overall fuel mass fraction in the previous section were derived for an
isothermal (i.e., non-vaporizing) spray. Evaporation complicates the problem further. Still, assuming
mixing-limited vaporization, only equilibrium considerations are needed to solve the problem.
For Le = 1, the law of energy conservation (in terms of relative enthalpy) has exactly the same
form as those for fuel mass and momentum. Therefore, without extra eort, the distribution of relative
enthalpy throughout the spray is obtained. When also Sc = 1, we have:
h(x, r) ha,
U(x, r)
=
= Y f (x, r) = Y f,l + Y f,v .
h f (T f,0 ) ha,
U0
(19.58)
For combustion modeling purposes the enthalpy eld needs to be converted to temperature, which
requires knowledge on the distribution of fuel over the (vapor and liquid) phases. Assuming mixinglimited evaporation, the distribution of fuel over the phases can be readily computed.
If Y f is known everywhere, Eq. (19.58) provides a relation between T (x, r) and Y f,v . Inserting
h = Y f,l h f,l + Y f,v h f,v + (1 Y f )ha and h f,v (T ) = h f,l (T ) + hvap (T ) it can be rewritten as
Y f,v =
(1 Y f )[ha (T a ) ha (T )] Y f [h f,l (T ) h f,l (T f,0 )]

.
hvap
(19.59)

An extra relation between Y f,v and T follows from the saturation condition:
Mf
(1 Y f ) 1
Y f,v =
Ma
1
X f,v
383
(19.60)
where X f,v is the saturated molar fuel fraction in the gas phase (notice the dierence with Y f,v which
is the fuel vapor mass fraction in the mixture). In an ideal gas approximation, X f,v = p s (T )/pa . In
Ref. [16] it is shown how this changes in a real gas context. An easier formulation of the latter, in
terms of enhancement factors, can be found in Ref. [45].
As in the Versaevel model, division of fuel over vapor and liquid phases is only relevant for the
upstream/inner part of the spray. There, the above two equations are both needed to solve for Y f,v and
Y f,l . More downstream (and, in this case, also further to the spray edges), only fuel vapor is present.
At the border of those regions, all liquid fuel has just evaporated. This contour is characterized by a
temperature T evap and by one particular mass fraction Y f,evap which in that case equals Y f and Y f,v .
Inserting this into Eq. (19.59) gives
Y f,evap
ha (T a ) ha (T evap )
=
,
1 Y f,evap h f,l (T evap ) + hvap (T evap ) h f,l (T 0 )
(19.61)
which can be proven to be equivalent to B in Eq. (19.33) and to 1/ in the Versaevel model, with the
same saturation temperature T evap = T s .
One issue not yet addressed above, is the eect of evaporation on density (via temperature). As
we saw in the Versaevel model, this does aect spray penetration. For a vaporizing spray, the local
density is given by
1
= Y
(19.62)
Y f,v
Ya
f,l
f,l + f,v + a
where all parameters depend on x and r. In this way, a coupling exists with the penetration law derived
before: evaporation aects density, which in turns aects the distribution of U/U0 and Y f . In Ref. [16]
this is stepped over rather loosely, making use of a simple scaling model for Y f,axis .
In a later paper [55] this coupling is accounted for in a self-consistent way. Consequently, the
model is no longer analytically solvable (mainly since the coupling between Y f,v and is non-linear),
and a numerical approach is used. This opens the way to take into account transient behavior (and
deal with time-variant boundary conditions) and the eect of non-unity Sc number. Yet the model is
still 1D, and therefore on the verge between phenomenological and CFD modeling.
An important advantage of the Valencia spray evaporation model is that it does not need any
adjustable constants in order to correlate experimental penetration and liquid length data. Still, it
remains questionable whether this also holds when comparing with other data sets than their own
often, best agreement is found with in-house data [75, 80, 18].
Combustion
Recently, Desantes et al. have extended the model of Ref. [55] to reacting sprays [17]. This is done
by interpreting the fuel mass fraction Y f as a mixture fraction during combustion. Combustion is
modeled using a one step irreversible reaction. Another change is in the state relationships, from
which local density and temperature are obtained. With combustion, these relations necessarily must
include the eect of chemical enthalpy (but including that is straightforward) and more species being
present.
384
Overall, the approach remains very much the same as for non-reacting sprays, which is elegant.
The work bears some similarity to that of Flynn et al. [25], who also use a physical spray model to
compute the temperature/equivalence ratio eld and then continue with combustion modeling in a
phenomenological fashion.
19.5
Recent developments and Outlook
In this section some points are briey sketched that are currently attracting attention.
Generally, there is a need for spray models that are capable to predict the penetration, evaporation
and combustion behavior of sprays, including the formation of emissions. The approach of using semiphenomenological models to provide source terms in CFD, thereby avoiding the inherent diculties
of Euler-Lagrangian models, seems to be the most promising.
Within the group of phenomenological models, the Valencia model seems the most promising.
It is certainly more physical (including radial variations) and more accurate than the Sandia-based
models, since no adjustable constants are needed to match the results with experimental data. On the
other hand, the Versaevel model has also proven its ability to provide a CFD code with reasonable
source terms. The CFD code rapidly translates the articially at source terms of the Versaevel
model into radial variations [10]. If the ame lift-o length is suciently large, including the radial
direction in the source term might not be needed. Yet, the Valencia model is expected to give better
radial distributions from the start. It must be investigated whether this makes a signicant dierence
in the combustion phase.
The above goal of accurate prediction of spray behavior should be accomplished over an extremely
wide range of (often multi-component) fuels, bringing about a variety of fuel properties and injection
conditions. Amongst these are extremely high pressures and various nozzle geometries. This is
directly related to the large variety of fuels that is claiming a role on the future fuel market. Examples
are biodiesel (with pure plant oils as an extreme case) and lighter molecules such as DME and ethanol.
Multi-component vaporization in sprays has been studied, amongst others, by Canaan et al. [12],
Siebers [74] and Myong et al. [51]. Usually it is found that steady liquid length is dominated by the
least volatile component in a mixture; more specically, it is found to correlate well with T 90 , the 90%
boiling point of the mixture. Recently, Pastor et al. have correlated liquid length results for biodiesel
blends in terms of mixing-limited vaporization [56].
Real gas eects in droplet-limited vaporization models were addressed, for instance, by Curtis et
al. [13] and Hohmann et al. [35]. Real gas eects in mixing-limited models were investigated recently
by Desantes et al. [16] and by Kurvers and Luijten [44, 45]. One aspect investigated in Ref. [45] is
the interaction between spray penetration and vaporization in the models of Refs. [75] and [80]. It
appears that mixture thermodynamics is decoupled from the spray penetration. The other way around,
it seems that vaporization does aect the spray penetration speed, as reported in Refs. [80, 55]. This
requires further investigation.
An extreme situation of real gas eects happens when the fuel enters the combustion chamber
under conditions close to the critical point. When the fuel-ambient mixture immediately reaches
the critical line, this results in ash boiling: the liquid phase instantly disappears by vigorous rupture. Although it might seem that every diesel spray would encounter supercritical conditions in the
cylinder, this is not the case. In an excellent review on supercritical droplet vaporization, Givler and
Abraham [26] show that the ambient pressure should be twice the fuels critical pressure and its temperature should be twice the fuels critical temperature, in order for the droplet to reach supercritical
conditions before complete vaporization.
385
The conditions under which the assumption of mixing-limited vaporization may be used, are not
yet established very well. An attempt in this direction by Iyer et al. [39] has not been fully convincing
yet, and is not really quantitative in terms of where the boundary of mixing-limited vaporization is. It
is recommended to investigate this transition by implementing the detailed droplet vaporization model
sketched in section 19.2.4 into, for instance, the Valencia penetration model.
References
[1] J. Abraham. What is adequate resolution in the numerical computations of transient jets? SAE
Tech. Paper 970051, 1997.
[2] J. Abraham and V. Magi. A Virtual Liquid Source (VLS) model for vaporizing diesel sprays.
SAE Tech. Paper 1999-01-0911, 1999.
[3] B. Abramzon and S. Sazhin. Convective vaporization of a fuel droplet with thermal radiation
absorption. Fuel, 85:3246, 2006.
[4] B. Abramzon and W.A. Sirignano. Droplet vaporization model for spray combustion calculations. Int. J. Heat Mass Transfer, 32(9):16051618, 1989.
[5] U. Aronsson, C. Chartier, O. Andersson, R. Egnell, J. Sj holm, M. Richter, and M. Ald n.

o
e
Analysis of the correlation between engine-out particulates and local in the lift-o region of a
heavy-duty diesel engine using raman spectroscopy. SAE Tech. Paper 2009-01-1357, 2009.
[6] J. Arr` gle, J.V. Pastor, and S. Ruiz. The inuence of injection parameters on diesel spray chare
acteristics. SAE Tech. Paper 1999-01-0200, 1999.
[7] R.S.G. Baert, P.J.M. Frijters, L.M.T. Somers, C.C.M. Luijten, and W.A. de Boer. Design and
operation of a high pressure, high temperature cell for HD diesel spray diagnostics: guidelines
and results. SAE Tech. Paper 2009-01-0649, 2009.
[8] C. Baumgarten. Mixture formation in internal combustion engines. Heat and Mass Transfer.
Springer, Berlin Heidelberg, 2006.
[9] J. Bayer and D.E. Foster. Zero-dimensional soot modeling. SAE Tech. Paper 2003-01-1070,
2003.
[10] C. Bekdemir. Numerical modeling of diesel spray formation and combustion. Technical Report
WVT 2008.15, Eindhoven University of Technology, 2008.
[11] M.D. Boot, C.C.M. Luijten, L.M.T. Somers, E.P. Rijk, and B.A. Albrecht. Spray impingement
in the early direct injection Premixed Charge Compression Ignition regime. SAE Tech. Paper
2010-01-1501, 2010.
[12] R.E. Canaan, J.E. Dec, G.M. Green, and D.T. Daly. The inuence of fuel volatility on the
liquid-phase fuel penetration in a heavy-duty D.I. diesel engine. SAE Tech. Paper 980510, 1998.
[13] E.W. Curtis, A. Uludogan, and R.D. Reitz. A new high pressure droplet vaporization model for
diesel engine modeling. SAE Tech. Paper 952431, 1995.
[14] J.E. Dec. A conceptual model of DI diesel combustion based on laser-sheet imaging. SAE Tech.
Paper 970873, 1997.
386
[15] J.M. Desantes, J. Arr` gle, J.V. Pastor, and A. Delage. Inuence of the fuel characteristics on the
e
injection process in a D.I. diesel engine. SAE Tech. Paper 980802, 1998.
[16] J.M. Desantes, J.J. L pez, J.M. Garca, and J.M. Pastor. Evaporative diesel spray modeling.
o
Atomization and Sprays, 17:193231, 2007.

[17] J.M. Desantes, J.V. Pastor, J.M. Garca-Oliver, and J.M. Pastor. A 1D model for the description
of mixing-controlled reacting diesel sprays. Combust. Flame, 156:234249, 2009.

[18] J.M. Desantes, J.V. Pastor, R. Payri, and J.M. Pastor. Experimental characterization of internal
nozzle ow and diesel spray behavior. part II: evaporative conditions. Atomization and Sprays,
15:517543, 2005.
[19] J.M. Desantes, R. Payri, J.M. Garca, and F.J. Salvador. A contribution to the understanding of
isothermal diesel spray dynamics. Fuel, 86:10931101, 2007.

[20] J.M. Desantes, R. Payri, J.M. Pastor, and J. Gimeno. Experimental characterization of internal
nozzle ow and diesel spray behavior. part I: nonevaporative conditions. Atomization and Sprays,
15:489516, 2005.
[21] J.M. Desantes, R. Payri, F.J. Salvador, and A. Gil. Development and validation of a theoretical
model for diesel spray penetration. Fuel, 85:910917, 2006.
[22] J.M. Desantes, R. Payri, F.J. Salvador, and V. Soare. Study of the inuence of geometrical and
injection parameters on diesel spray characteristics in isothermal conditions. SAE Tech. Paper
2005-01-0913, 2005.
[23] S. di Stasio and L. Allocca. Inuence of the gas ambient nature on diesel spray properties
at high injection pressure: experimental results. In J. Whitelaw, F. Payri, and J.-M. Desantes,
editors, THIESEL 2000. Thermo- and Fluid-dynamic processes in Diesel engines, pages 95108,
Valencia, 2000. Springer.
[24] L.A. Dombrovsky and S.S. Sazhin. A parabolic temperature prole model for heating of
droplets. J. Heat Tranfers, 125:535537, 2003.
[25] P.F. Flynn, R.P. Durrett, G.L. Hunter, A.O. zur Loye, O.C. Akinyemi, J.E. Dec, and C.K. Westbrook. Diesel combustion: an integrated view combining laser diagnostics, chemical kinetics,
and empirical validation. SAE Tech. Paper 1999-01-0509, 1999.
[26] S.D. Givler and J. Abraham. Supercritical droplet vaporization and combustion studies. Prog.
Energy Combust. Sci., 22:128, 1996.
[27] J.B. Heywood. Internal combustion engine fundamentals. McGraw-Hill, New York, 1988.
[28] B.S. Higgins, C.J. Mueller, and D.L. Siebers. Measurements of fuel eects on liquid-phase
penetration in diesel sprays. SAE Tech. Paper 1999-01-0519, 1999.
[29] H. Hiroyasu. Diesel engine combustion and its modeling. In Proceedings of the Int. Symp. on
Diagnostics and Modeling of Combustion in Reciprocating Engines, pages 5375, Tokyo, 1985.
[30] H. Hiroyasu and M. Arai. Fuel spray penetration and spray angle in diesel engines. Trans. JSAE,
21(5), 1980.
387
[31] H. Hiroyasu and M. Arai. Structures of fuel sprays in diesel engines. SAE Tech. Paper 900475,
1990.
[32] H. Hiroyasu, M. Arai, and M. Tabata. Empirical equations for the Sauter Mean Diameter of a
diesel spray. SAE Tech. Paper 890464, 1989.
[33] H. Hiroyasu and T. Kadota. Development and use of a spray combustion modeling to predict
diesel engine eciency and pollutant emissions. SAE Trans., 85(760129), 1976.
[34] H. Hiroyasu, T. Kadota, and M. Arai. Development and use of a spray combustion modeling to
predict diesel engine eciency and pollutant emissions. Bulletin JSME, 26(214), 1983.
[35] S. Hohmann, M. Klingsporn, and U. Renz. An improved model to describe spray evaporation
under diesel-like conditions. SAE Tech. Paper 960630, 1996.
[36] S. Hohmann and U. Renz. Numerical simulation of fuel sprays at high ambient pressure: the
inuence of real gas eects and gas solubility on droplet vaporization. Int. J. Heat Mass Transfer,
46:30173028, 2003.
[37] G.L. Hubbard, V.E. Denny, and A.F. Mills. Droplet evaporation: eects of transients and variable
properties. Int. J. Heat Mass Transfer, 18:10031008, 1975.
[38] V.A. Iyer, J. Abraham, and V. Magi. Exploring injected droplet size eects on steady liquid
penetration in a diesel spray with a two-uid model. Int. J. Heat Mass Transfer, 45:519531,
2002.
[39] V.A. Iyer, S.L. Post, and J. Abraham. Is the liquid penetration in diesel sprays mixing controlled?
In Proceedings of the Combustion Institute, pages 11111118. The Combustion Institute, 2000.
[40] D. Jung and D.N. Assanis. Multi-zone DI diesel spray combustion model for cycle simulation
studies of engine performance and emissions. SAE Tech. Paper 2001-011246, 2001.
[41] G. Kalghatgi. Lift-o heights and visible lengths of vertical turbulent jet diusion ames in still
air. Combust. Sci. Techn., 41:1729, 1984.
[42] J. Kang, C. Bae, and K.O. Lee. Initial development of non-evaporating diesel sprays in commonrail injection systems. Int. J. Engine Res., 4(4):283298, 2003.
[43] R.J.H. Klein-Douwel, P.J.M. Frijters, L.M.T. Somers, W.A. de Boer, and R.S.G. Baert. Macroscopic diesel fuel spray shadowgraphy using high speed digital imaging in a high pressure cell.
Fuel, 86(12-13):19942007, 2007.
[44] C. Kurvers and C.C.M. Luijten. Real gas eects in siebers mixing-limited spray vaporization
model. SAE Tech. Paper 2010-01-1497, 2010.
[45] C.C.M. Luijten and C. Kurvers. Real gas eects in mixing-limited spray modeling. Atomization
and Sprays, page (accepted), 2010.
[46] R. Meyer and J.B. Heywood. Evaporation of in-cylinder liquid fuel droplets in an SI engine: a
diagnostic-based modeling study. SAE Tech. Paper 1999-01-0567, 1999.
[47] S. Moon, Y. Matsumoto, and K. Nishida. Entrainment, evaporation and mixing characteristics
of diesel sprays around end-of-injection. SAE Tech. Paper 2009-01-0849, 2009.
388
[48] C.J. Mueller, L.M. Pickett, D.L. Siebers, W.J. Pitz, C.K. Westbrook, and G.C. Martin. Eects
of oxygenates on soot processes in DI diesel engines: experimental and numerical simulations.
SAE Tech. Paper 2003-01-1791, 2003.
[49] M.P. Musculus, J.E. Dec, and D.R. Tree. Eects of fuel parameters and diusion ame lift-o
on soot formation in a heavy-duty DI diesel engine. SAE Tech. Paper 2002-01-0889, 2002.
[50] M.P. B. Musculus, T. Lachau, L.M. Pickett, and C.A. Idicheria. End-of-injection over-mixing
and unburned hydrocarbon emissions in low-temperature-combustion diesel engines. SAE Tech.
Paper 2007-01-0907, 2007.
[51] K. Myong, M. Arai, H. Suzuki, J. Senda, and H. Fujimoto. Vaporization characteristics and
liquid-phase penetration for multi-component fuels. SAE Tech. Paper 2004-01-0529, 2004.
[52] J.D. Naber and D.L. Siebers. Eects of gas density and vaporization on penetration and dispersion of diesel sprays. SAE Tech. Paper 960034, 1996.
[53] W. Ohnesorge. Die Bildung von Tropfen an D sen und die Au sung ssiger Strahlen.
u
o
u
Zeitschrift f r angewandte Mathematik und Mechanik, 16(6):355358, 1931.
u
[54] J.V. Pastor, J.M. Garca, J.M. Pastor, and L.D. Zapata. Evaporating diesel spray visualization
using a double-pass shadowgraphy/schlieren imaging. SAE Tech. Paper 2007-24-0026, 2007.

[55] J.V. Pastor, J.J. L pez, J.M. Garca, and J.M. Pastor. A 1D model for the description of mixingo
controlled inert diesel sprays. Fuel, 87:28712885, 2008.

[56] J.V. Pastor, J.M. Pastor, J. Gimeno, and J.-G. Nerva. The eect of biodiesel fuel blend rate on
the liquid-phase fuel penetration in diesel engine conditions. SAE Tech. Paper 2009-24-0051,
2009.
[57] F. Payri, J. Arr` gle, J.J. L pez, and S. Hermens. Eect of cavitation on the nozzle outlet ow,
e
o
spray and ame formation in a diesel engine. SAE Tech. Paper 2006-01-1391, 2006.
[58] R. Payri, J.M. Garca, F.J. Salvador, and J. Gimeno. Using spray momentum ux measurements
to understand the inuence of diesel nozzle geometry on spray characteristics. Fuel, 84:551561,
2005.
[59] R. Payri, F.J. Salvador, J. Gimeno, and L.D. Zapata. Diesel nozzle geometry inuence on spray
liquid-phase fuel penetration in evaporative conditions. Fuel, 87:11651176, 2008.
[60] N. Peters. Turbulent combustion. Cambridge University Press, Cambridge, 2000.
[61] L.M. Pickett, S. Kook, and T.C. Williams. Visualization of diesel spray penetration, cool-ame,
ignition, high-temperature combustion, and soot formation using high-speed imaging. SAE Tech.
Paper 2009-01-0658, 2009.
[62] L.M. Pickett and D.L. Siebers. Fuel eects on soot processes of fuel jets at DI diesel conditions.
SAE Tech. Paper 2003-01-3080, 2003.
[63] L.M. Pickett and D.L. Siebers. Soot in diesel fuel jets: eects of ambient temperature, ambient
density, and injection pressure. Combust. Flame, 138:114135, 2004.
389
[64] L.M. Pickett, D.L. Siebers, and C.A. Idicheria. Relationship between ignition processes and the
lift-o length of diesel fuel jets. SAE Tech. Paper 2005-01-3843, 2005.
[65] L. Postrioti and S. Ubertini. An integrated experimental-numerical study of hsdi diesel injection
system and spray dynamics. SAE Tech. Paper 2006-01-1389, 2006.
[66] R.D. Reitz and F.V. Bracco. On the dependence of spray angle and other spray parameters on
nozzle design and operating conditions. SAE Tech. Paper 790494, 1979.
[67] R.D. Reitz and F.V. Bracco. Mechanisms of break-up of round liquid jets. Encyclopedia of Fluid
Mechanics, 3:233249, 1986.
[68] G. Rudinger. Fundamentals of gas-particle ow, volume 2 of Handbook of Powder Technology.
Elsevier, Amsterdam, 1980.
[69] S.S. Sazhin. Advanced models of fuel droplet heating and evaporation. Prog. Energy Combust.
Sci., 32:162214, 2006.
[70] X.L.J. Seykens, L.M.T. Somers, and R.S.G. Baert. Detailed modeling of common rail fuel
injection process. MECCA, III(2+3):3039, 2005.
[71] B. Higgins D. Siebers. Measurements of the ame lift-o location on DI diesel sprays using OH
chemiluminescence. SAE Tech. Paper 2001-01-0918, 2001.
[72] D. Siebers and B. Higgins. Flame lift-o on direct-injection diesel sprays under quiescent conditions. SAE Tech. Paper 2001-01-0530, 2001.
[73] D. Siebers, B. Higgins, and L. Pickett. Flame lift-o on direct-injection diesel fuel jets: oxygen
concentration eects. SAE Tech. Paper 2002-01-0890, 2002.
[74] D.L. Siebers. Liquid-phase fuel penetration in diesel sprays. SAE Tech. Paper 980809, 1998.
[75] D.L. Siebers. Scaling liquid-phase fuel penetration in diesel sprays based on mixing-limited
vaporization. SAE Tech. Paper 1999-01-0528, 1999.
[76] G.J. Smallwood and O.L. G lder. Views on the structure of transient diesel sprays. Atomization
u
and Sprays, 10:355386, 2000.
[77] G. Stiesch. Modeling engine spray and combustion process. Heat and Mass Transfer. Springer,
Berlin Heidelberg, 2003.
[78] S.R. Turns. An introduction to combustion: concepts and applications. McGraw-Hill, New
York, 1996.
[79] D. Verhoeven, J.L. Vanhemelryck, and T. Baritaud. Macroscopic and ignition characteristics of
high-pressure sprays of single-component fuels. SAE Tech. Paper 981069, 1998.
[80] P. Versaevel, P. Motte, and K. Wieser. A new 3D model for vaporizing diesel sprays based on
mixing-limited vaporization. SAE Tech. Paper 2000-01-0949, 2000.
[81] Y. Wan and N. Peters. Scaling of spray penetration with evaporation. Atomization and Sprays,
9:111132, 1999.
[82] T. Yoshizaki, K. Nishida, and H. Hiroyasu. Approach to low NO x and smoke emission engines
by using phenomenological simulation. SAE Tech. Paper 930612, 1993.
390
Part 6
Applications
391
Chapter 20
Thermo-acoustics in laminar ames

V.N. Kornilov
393
394
Chapter 20 V.N. Kornilov
Chapter 21
Thermo-acoustics in turbulent ames

J.B.W. Kok
395
396
Chapter 21 J.B.W. Kok
Chapter 22
Gas turbines
J.B.W. Kok
397
398
Chapter 22 J.B.W. Kok
Chapter 23
Fire
B. Merci
23.1
Introduction
Fire is a complex natural phenomenon, in which many physical and chemical processes strongly
interact. From a combustion point of view, re is essentially (in most cases) an uncontrolled form of
non-premixed combustion. This will be explained below.
In order to reduce the risk and consequences of re, which is often considered an undesired phenomenon, engineering principles can be used as tool. This multi-disciplinary approach to re protection is called Fire Safety Engineering (FSE). Apart from combustion, essential aspects in FSE are
uid mechanics (turbulence), heat transfer, structural stability, explosion safety, risk management, human behaviour, detection, suppression, etc. Clearly, an entire educational programme can be devoted
to FSE and indeed, Ghent University (Prof. Bart Merci) coordinates a 2-year International Master of
Science in Fire Safety Engineering programme (http://www.imfse.ugent.be), with the University of
Edinburgh (Scotland) and Lund University (Sweden) as partners.
In the text below, we will focus on some combustion (and heat transfer) aspects of re only. Most
of the information stems from basic text books [1-5]. We will focus on enclosure res, not wildland
res, as enclosure res are the most common in our surroundings. We will not consider the problem
of ignition, which is a very complex study on itself.
We will also only focus on res with aming combustion, not smouldering res. This implies
that the combustible material is transformed into gaseous fuel rst and that this gaseous fuel then
reacts with oxygen in the air to form ames.
23.2
Fire triangle
In order to have a sustained re on our planet, three ingredients are required, from a physical point of
view: fuel, oxygen and heat. The rst two ingredients are typically always there, so that, as soon as
there is enough heat, there is a risk for re. This concept is called the re triangle.
Slightly more precisely, in the context of res with aming combustion, a mixture of gaseous
fuel and oxygen, within the so-called ammability limits, must nd a region of suciently high
temperature (and chemical radicals) for chemical reaction (combustion) to be initiated.
The gaseous fuel is generated by evaporation, in case of a liquid fuel, or pyrolysis, in case of a
solid combustible material. Both processes are globally endothermic, i.e. they require energy. This
energy stems from the ames in the gas phase. Consequently, a closed loop is established. If the heat
transfer from the ames is not sucient to maintain the evaporation / pyrolysis process, the re will
extinguish. If the heat transfer rate is excessive, compared to what is required by the evaporation /
399
400
Chapter 23 B. Merci
pyrolysis, the re will grow and/or become more severe. If there is a balance, a steady state situation
is met. This closed loop is an essential dierence from controlled combustion, where the thermal
power of the device is controlled by the mass ow rate of fuel (and oxygen). In case of re with
aming combustion, the fuel mass ow rate is strongly aected by the re heat release rate (HRR)
itself.
23.3
23.3.1
Fire plumes
Jet ame versus natural re ame
As mentioned, in case of re with aming combustion, fuel and oxidiser (oxygen from air) are initially
separate and burn as they mix. They burn wherever the concentrations of fuel and oxidiser are within
(high temperature) ammability limits. The appearance of the ame depends on the nature of the fuel
and on the velocity (momentum) of the fuel jet with respect to the surrounding air.
The dierence between jet ames, which have a relatively high momentum, and natural re
ames, which are driven by buoyancy driven, can be quantied by means of the Froude number Fr,
which expresses the relative importance of inertia and buoyancy in the system under consideration:
Fr =
U 2
gL
(23.1)
In this expression, is the density, U is the characteristic velocity (at the inlet of the fuel), =
amb is the dierence between the local density and the density of the ambient environment, g is
the gravity acceleration constant and L is a characteristic length. Although buoyancy is an upward
phenomenon, typically the pool diameter D is chosen as characteristic length. The ration is also
typically not taken into account, so that the following denition of Fr is used:
Fr =
U2
gD
(23.2)
In natural re ames, Fr < 1, as seen in gure 23.1.

Another way to characterise a re in a non-dimensional manner, which is typically used in the re
community (see gure 23.1), is through the non-dimensional re HRR:
Q =
Q
amb c p T amb (gD)1/2 D2
(23.3)
In this expression, Q is the re HRR, c p is the specic heat of air and T amb is ambient air temperature.
As the re size increases, the value of Q typically decreases (gure 23.2).
23.3.2
Flame height
The ame height can be dened by means of an intermittency factor I. On a position where there is
always a ame, I = 1. When there is never a ame, i.e. suciently high above the re source, I = 0.
The mean ame height can be dened as the position / height where I = 0.5 (gure 23.3). The ame
height can be correlated to the re source diameter and HRR, but we do not go into details here.
Fire
401
Figure 23.1: Natural re ames (I, II, III) and jet ames (V).
Figure 23.2: Illustration of relation of Q to pool re diameter.
402
Chapter 23 B. Merci
Figure 23.3: Concept of intermittency factor I to determine the ame height.
23.3.3
Fire plume
Using the notion of intermittency, the re plume can be subdivided into three regions (gure 23.4).
In the near eld region, just above the burner / re fuel surface, there is a persistent ame
(I = 1). In this region, there is an accelerating ow of burning gases (gure 23.5) and the temperature
is nearly constant (gure 23.6), typically around 11001200 K. In the buoyant plume region (I = 0),
the upward velocity and temperature decrease with height (gures 23.5 and 23.6), due to entrainment
of fresh air into the rising plume. This zone is also called the smoke plume. Smoke is to be seen
as hot gases (combustion products), with particles. In between the persistent ame region and the
buoyant plume region, is the intermittent region, with intermittent aming. In this zone, the velocity
is practically constant (gure 23.5).
The entrainment of fresh air into the re plume is dierent for each of the three regions and specic
correlations, particularly the ones developed by Heskestad, account for these dierences, be it on an
empirical basis. We will not go into more detail here on this correlations.
23.4
Enclosure re
An essential aspect of enclosure res is the interaction of the re and smoke with compartment boundaries.
Fire
403
Figure 23.4: Three regions of a re plume.
An important issue in enclosure res is that there can be a lack of ventilation openings. The re
can then, at a certain stage, be ventilation controlled rather than fuel controlled. We will not go
into more detail here.
Another issue is that if the re ames are close to a wall, the presence of the wall will eectively
modify the mixing process of combustible gases and oxygen, primarily due to a blocking eect.
Consequently, the ame shape and temperature will change and correlations for ame heights of free
res are not valid.
Unless the ceiling is very high, the most important feature of an enclosure re (with sucient
ventilation openings) is the interaction with the ceiling. The rising re plume impinges onto the
ceiling and the vertical motion is diverted into a horizontal motion. The ceiling jet phenomenon is
discussed in some more detail below. There is typically also the build-up of a smoke layer.
23.4.1
Smoke layer
A possible situation in steady re conditions is the existence of a hot smoke layer under the ceiling, which is relatively clearly separated from a cold bottom layer (gure 23.7). The zone model
approach to simulate enclosure res relies on the existence of these two zones, which have uniform
temperature per zone. Obviously, this will never truly be the case in reality, but under certain circumstances it can be an acceptable approximation.
404
Chapter 23 B. Merci
Figure 23.5: Upward velocity in the three regions of a re plume.
The smoke layer is built from the ceiling and descends, while its temperature varies in time. The
steady position, if a steady situation is met at all, depends on the ventilation conditions and the re
source primarily. We will not go into more detail here.
23.4.2
Ceiling jet
An interesting phenomenon, also from a combustion point of view, is the ceiling jet phenomenon.
As mentioned, due to the blocking eect of the ceiling, the rising re plume is deected from an
upward motion to a horizontal motion under the ceiling. This is certainly the case when the smoke
layer has not been built yet.
If the ames are relatively low, compared to the ceiling height, it is the smoke plume that is
impinging onto the ceiling. The radially outward horizontal smoke motion is relevant for the determination of the time for the smoke to reach a detector, mounted on the ceiling. This will depend on the
re source, which aects the upward re plume momentum. It is noteworthy that entrainment of fresh
air into the horizontally moving smoke is by far less ecient than in the rising plume. The smoke is
indeed hotter than the underlying fresh air, which is a stable situation from a uid mechanics point
of view, so that mixing is not stimulated.
If the ames reach the ceiling, there on must distinguish between a situation where only the ame
tips reach the ceiling and the situation where a larger body of ames interacts with the ceiling. In the
former case, the burnt gases are fuel lean and thus the less ecient entrainment of fresh air into the
hot gases is not very important. In the latter case, on the other hand, the burning gases are typically
fuel rich and a considerable ame extension is possible due to the attenuated mixing with fresh air
(gure 23.8).
Fire
405
Figure 23.6: Temperature in the three regions of a re plume.
23.5
Flame spread
23.5.1
Heat transfer
Heat transfer is essential in a re. Heat is required to produce combustible gases, either through
evaporation of a liquid fuel or through pyrolysis of a solid fuel. This is discussed in the next section.
In the gas phase, convection and radiation are the two primary mechanisms of heat transfer. For a
free re, the amount of radiation from the ames depends on the fuel type. For alcohols, e.g., which
produce very small amounts of soot, the radiative fraction is close to zero. In very sooty fuel, can
exceed 50%. As a rough estimate, the typical value for is 0.2 to 0.3. This implies that only 80 or
70% is then given as convective heat to generate the upward, buoyancy driven, plume motion. The
rest of the re HRR is, in this sense, lost as radiation.
In enclosure res, the situation is less straightforward. In the compartment boundaries (oor,
walls, ceiling), conduction is an important mechanism. This way, heat is removed from the enclosure.
The (kc) value of the material (with k the conduction coecient of the solid material) is crucial:
low values imply that little heat is removed as the material goes in thermal equilibrium with its environment rapidly, while high values have the opposite behaviour. Through openings (windows, door
openings, ceiling vents), radiation is lost from the enclosure through convection and/or radiation.
When a smoke layer is present, heat is exchanged with the ceiling and part of the walls (in contact
with the smoke layer) by means of convection, but the layer also radiates to the rest of the enclosure,
typically also onto the re source. The ceiling, walls and oor also all radiate towards each other. Note
406
Chapter 23 B. Merci
Figure 23.7: Schematic representation of hot upper and cold bottom layer in an enclosure re.
Figure 23.8: Schematic representation of ame extension under the ceiling in an enclosure re.
Fire
407
that, in an enclosure, the radiative fraction of the ames may be lost with respect to the upward
buoyant plume region, but not lost from the enclosure.
23.5.2
Pyrolysis
As already mentioned, for combustible gases to be generated, the liquid fuel must evaporate or the
solid fuel must degradate. The latter process is called pyrolysis. In case of re with aming combustion, this is high temperature pyrolysis. From now on, we will focus on this type.
Pyrolysis is globally an endothermic process, using the heat from the ames in the gas phase.
This heat is transferred by convection and radiation. Pyrolysis is essentially a chemical process: the
structure of the solid changes and combustible gases (volatiles) are released. The solid can leave a
char layer (charring material) upon pyrolysis or not (non-charring materials). When a char layer
is left as residue, an oxidation process can take place, but in case of re, this is not very relevant in
many cases. This is a strong dierence from controlled solid fuel combustion.
In modelling of pyrolysis, the chemistry is strongly simplied, either to an Arrhenius expression
or even to innitely fast chemistry. In the latter case, the pyrolysis region is reduced to an innitely
thin front, at pyrolysis temperature. Thus, the process is eectively reduced to a physical, rather than
chemical, representation of the phenomenon. Considering typical time scales, this is often justied. A
recently developed pyrolysis model of this type, applicable to dry and wet charring and non-charring
materials, is described in [6]. In essence, the energy equation is solved on a xed mehs and the motion
of the pyrolysis front is derived from a local heat balance around the front. The volatiles released are
assumed in thermal equilibrium with the solid material. We will not go into more detail here.
23.5.3
Types of ame spread
Before going into the matter of ame spread over solid fuel surfaces, we recall the fundamental mechanism, as represented in gure 23.9. Part of the heat from the re HRR is transferred back to the re
source surface by radiation and convection. The rest is radiated to the surroundings and transferred
into the re plume and thus convected away from the re source. The incoming heat (onto the re
source) generates the mass ow rate of combustible gases (by evaporation or pyrolysis), and is partly
lost from the re source (by conduction into the fuel and radiation (and convection) from the fuel
surface into the surroundings).
Now we consider ame spread over surfaces. The basic principle remains the same: heat from the
ames is used for the pyrolysis process, by which combustible gases (volatiles) are produced. These
mix with air and react with the oxygen, leading to heat from the combustion. If this heat release
rate exceeds what is necessary for the pyrolysis process, the ame spread will accelerate. If it is
insucient, it will slow down and possibly extinguish. A steady situation is also possible.
An important issue is the ow of air, compared to the direction of the ame spread. When they
are in the same direction, it is called co-ow ame spread or concurrent ame spread. In case of a
vertically oriented surface, this is the natural situation for upward ame spread: buoyancy induces an
upward air ow near the surface. As a consequence, the ames are driven towards the virgin solid
material and the view factor for radiation is high. This implies an intense heat transfer and rapid ame
spread. This is shown in gure 23.10 (bottom). Note that convective heat transfer towards the virgin
material also occurs, but is often less important.
In case of counter-current ame spread (gure 23.10, top), the ame is driven towards the material
that has already pyrolysed. Consequently, the radiative heat transfer in the gas phase towards the
virgin material is far less eective than in concurrent ame spread, due to a much lower view factor.
408
Chapter 23 B. Merci
Figure 23.9: Schematic representation of heat and mass transfer in the case of a steady pool re.
Fire
409
Indirectly, the heat is transferred in the solid material by conduction and there is also some convective
heat transfer in the gas phase, but clearly this is all less intense than for concurrent ame spread.
In vertically oriented surfaces, counter-current ame spread corresponds to downward ame spread:
buoyancy again induces an upward ow.
Flame spread can be simulated by coupling of CFD, including gas phase combustion, to pyrolysis
modelling inside the solid material. This is an area of ongoing research.
References
[1] G. Cox and J. Whitelaw, Comubstion Fundamentals of Fire, Academic Press (1995).
[2] D. Drysdale, An Introduction to Fire Dynamics, 2nd Ed., John Wiley and Sons Ltd. (1998).
[3] B. Karlsson and J. Quintiere, Enclosure Fire Dynamics, CRC Press (2000).
[4] J. G. Quintiere, Fundamentals of Fire Phenomena, John Wiley and Sons Ltd. (2006).
[5] SFPE Handbook of Fire Protection Engineering, 3rd Ed., NFPA (2007).
[6] S. Wasan, P. Rauwoens, J. Vierendeels and B. Merci, An Enthalpy-Based Pyrolysis Model for
Charring and Non-Charring Materials in Case of Fire, Combustion and Flame, Vol. 157, pp.
715-734 (2010).
410
Chapter 23 B. Merci
Figure 23.10: Schematic representation of concurrent (bottom) and counter-current (top) ame spread over a
solid surface.
Chapter 24
Radiative heat transfer

P.J. Coelho
411

JMBC2010

Загружено:

Сведения о документе

Исходное описание:

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

JMBC2010

Загружено:

Авторское право:

Доступные форматы

Lecture notes of the

September 28 October 1, 2010

prof.dr.ir. D.J.E.M. Roekaerts

Introduction to Combustion L.P.H. de Goey

Governing equations L.M.T. Somers

Premixed laminar ames L.P.H. de Goey

Numerical Modelling of Laminar Flames J.H.M. ten Thije Boonkkamp

Reduced chemical models L.M.T. Somers

Introduction to turbulent combustion D.J.E.M. Roekaerts

Non-premixed amelet methods L.M.T. Somers

Premixed amelet methods R.J.M. Bastiaans

Probability Density Function (PDF) methods D.J.E.M. Roekaerts

An introduction to laser diagnostics N.J. Dam

Recent developments in laser diagnostics M. Ald n

LDA and PIV M.J. Tummers en L.F.G. Geers

Thermometry by coherent anti-Stokes Raman spectroscopy E.H. van Veen

Application of laser diagnostics Th.H. van der Meer

Clean Combustion Concepts

Preheated dilute combustion D.J.E.M. Roekaerts and M.J. Tummers

Clean combustion concepts in furnaces E.-S. Cho, B. Danon and W. de Jong

New combustion concepts in Diesel engines L.M.T. Somers

Thermo-acoustics in laminar ames V.N. Kornilov

Thermo-acoustics in turbulent ames J.B.W. Kok

Gas turbines J.B.W. Kok

Radiative heat transfer P.J. Coelho

Prof.dr. M. Ald n (Marcus) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lund Institute of Technology

Dr.ir. W. de Jong (Wiebren) . . . . . . . . . . . . . . . . Delft University of Technology

Dr.ir. J.A. van Oijen (Jeroen) . . . . . . . . . . . . . . . . . . Eindhoven University of Technology

Chapter 1 L.P.H. de Goey

Figure 1.1: The modied Borghi-diagram for premixed turbulent combustion.

Flame fronts and burning velocity

Chapter 1 L.P.H. de Goey

Figure 1.4: Regimes in non-premixed turbulent combustion.

Chapter 1 L.P.H. de Goey

Figure 1.5: Mixing of two streams with fuel and oxidiser

Structure of non-premixed ames

sYF Yox + Yox,2

Chapter 1 L.P.H. de Goey

(Vi Yi ) = i , (i = 1, ..., N).

specic internal energy e = h p/,

Chapter 2 L.M.T. Somers

or for that matter the conservation of specic total energy, e = e + v2 /2,

Chapter 2 L.M.T. Somers

There appears to be some discussion lately on whether the

should not read

Simplied models for the diusive ux

X3 [(M3 /M2 )D13 D12 ]

and similar relations for D21 , D23 , D32 , D13 , D31 .

Chapter 2 L.M.T. Somers

Or in a binary mixture, but that is trivial.

Evaluating thermodynamic and transport properties

Chapter 2 L.M.T. Somers

and complete the mixture averaged thermodynamical and transport properties.

with R the universal gas constant.

an,i (ln T )n1

bn,i (ln T )n1

dn,i j (ln T )n1

remaining part of this lecture.

Chemical aspects in combustion

General elementary reactions

In general, any elementary reaction can be written in the form,

Chapter 2 L.M.T. Somers

where the symbol