Sunteți pe pagina 1din 29

Combustion and Flame 155 (2008) 380–408

www.elsevier.com/locate/combustflame

Formulation reproducing the ignition delays simulated by a


detailed mechanism: Application to n-heptane combustion
Bruno Imbert a,∗ , Fabien Lafosse b , Laurent Catoire b ,
Claude-Étienne Paillard b , Boris Khasainov c
a LENI, IGM, STI, EPFL, Station 9, CH-1015 Lausanne, Switzerland
b ICARE, CNRS, and University of Orléans, 1C Av. de la Recherche Scientifique, F-45067 Orléans, France
c LCD, CNRS, ENSMA, 1 Av. Clément Ader, F-86961 Futuroscope, France

Received 9 August 2007; received in revised form 30 January 2008; accepted 7 May 2008

Abstract
This article is part of the project to model the kinetics of high-temperature combustions, occurring behind
shock waves and in detonation waves. The “conventional” semi-empirical correlations of ignition delays have
been reformulated, by keeping the Arrhenius equation form. It is shown how a polynomial with 3N coefficients
(where N ∈ [1, 4] is the number of adjustable kinetic parameters, likely to be simultaneously chosen among the
temperature T , the pressure P , the inert fraction X Ar , and the equivalence ratio Φ) can reproduce the delays
predicted by the Curran et al. [H.J. Curran, P. Gaffuri, W.J. Pitz, C.K. Westbrook, Combust. Flame 129 (2002)
253–280] detailed mechanism (565 species and 2538 reactions), over a wide range of conditions (comparable with
the validity domain). The deviations between the simulated times and their fits (typically 1%) are definitely lower
than the uncertainties related to the mechanism (at least 25%). In addition, using this new formalism to evaluate
these durations is about 106 times faster than simulating them with S ENKIN (C HEMKIN III package) and only 10
times slower than using the classical correlations. The adaptation of the traditional method for predicting delays is
interesting for modeling, because those performances are difficult to obtain simultaneously with other reduction
methods (either purely mathematical, chemical, or even mixed). After a physical and mathematical justification
of the proposed formalism, some of its potentialities for n-heptane combustion are presented. In particular, the
trends of simulated delays and activation energies are shown for T ∈ [1500 K, 1900 K], P ∈ [10 kPa, 1 MPa],
XAr ∈ [0, 0.7], and Φ ∈ [0.25, 4.0].
© 2008 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

Keywords: Kinetics; Mechanism reduction; Ignition delays; Modeling; Detonation; Shock; n-Heptane

1. Introduction

1.1. General context: n-Heptane autoignition and


detonation

* Corresponding author. Fax: +41 (0) 21 693 35 02. Previous articles [1–4] have aimed at providing a
E-mail address: bruno.imbert@epfl.ch (B. Imbert). kinetic database able to contribute to the choice of a
0010-2180/$ – see front matter © 2008 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.combustflame.2008.05.011
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 381

Nomenclature

Subscripts LT reduced variable related to the logarithm


1 initial state of the driven section gas of the temperature T (Section 3.6) (di-
5 state of the driven section gas behind the mensionless)
reflected shock LX reduced variable related to the logarithm
CJ established according to the Chapman– of the inert fraction X Ar (Section 3.6)
Jouguet model (dimensionless)
cor related to a correlated term (index some- LΦ reduced variable related to the logarithm
times omitted to shorten the expression) of the equivalence ratio Φ (Section 3.6)
exp associated to an experimental measure- (dimensionless)
ment Lτ reduced variable related to the logarithm
i related to induction (process similar to of the ignition delay τ (dimensionless)
autoignition in the case of a detonation) M number of conditions used for the kinetic
min minimum database (dimensionless)
max maximum
M matrix (dimensionless)
opt optimum (minimum or maximum)
MY molar mass of species Y (g mol−1 )
Reac linked to the reactive part (fuel and oxy-
n unspecified integer (dimensionless)
gen) of a flammable mixture
ref related to a reference state: N number of adjustable parameters (di-
T ref = 1700 K, P ref = 100 kPa, Φ ref = mensionless)
1.0 and X Ar, ref = 0 p pressure coefficient (dimensionless)
sim associated to the predictions simulated P pressure (bar or kPa)
by a kinetic mechanism (here detailed) R = 8.314 gas constant (J mol−1 K−1 )
vN bounded to the average conditions be- T temperature (K)
hind the detonation front, according to x implicit dilution coefficient (dimension-
the ZND model less)
Variables (Latin and Greek letters) XY molar fraction (or percentage) of species
Y (argon, if Y not expressed with X in
[Y] molar concentration of species Y
index) (dimensionless or %)
(mol m−3 )
u particle velocity expressed in the refer-
A pre-exponential factor (s m3 · (α+β+γ )
ential related to the shock wave front
mol−(α+β+γ ) )
D detonation celerity (m s−1 ) (m s−1 )
Da Damköhler number, i.e., the ratio of V vector (dimensionless)
the speeds related to the chemical pro- α partial pseudo-order related to n-heptane
cesses and the mass transfers (dimen- (dimensionless)
sionless) β partial pseudo-order related to oxygen
Ea activation energy (kJ mol−1 ) (dimensionless)
Gτcor standard geometrical deviation between γ partial pseudo-order related to argon (di-
a set of reference simulated delays τ sim , mensionless)
and times τ cor correlating them (Ap- i induction distance (m)
pendix A.4) (dimensionless) τcor standard (arithmetic) deviation between
i index related to LP (dimensionless) a set of reference simulated delays τ sim ,
IT reduced variable related to the inverse of and times τ cor correlating them (dimen-
the temperature T (Section 3.6) (dimen- sionless)
sionless)
Φ equivalence ratio of the flammable mix-
j index related to LX (dimensionless)
ture (dimensionless)
k index related to LΦ (dimensionless)
λ average cell width of a self-sustained
l index related to one condition (thermo-
dynamic state and composition) (dimen- detonation (mm)
sionless) θa activation temperature (K)
LP reduced variable related to the logarithm ρ density (kg m−3 )
of the pressure P (Section 3.6) (dimen- τ ignition delay (or induction time in the
sionless) detonation case) (s)
382 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

Acronyms and abbreviations lg decimal logarithm ⇔ lg(10) = 1


CFD computational fluid dynamics ln neperian logarithm ⇔ ln(e) = 1
NTC negative temperature coefficient
CJ Chapman–Jouguet [8,9]: “0D” thermo-
PDE pulsed detonation engine
dynamic detonation model
QSSA quasi-steady state approximation
CSP computational singular perturbation RRKM Rice–Ramsperger–Kassel–Marcus [32]:
fy (z) multivariable function, only dependent model of chemical reactivity
on z by fixing the parameter y ZND Zel’dovitch–von Neumann–Döring [10–
ILDM intrinsic low-dimensional manifold 12]: “0D” kinetic detonation model

liquid fuel appropriate to the pulsed detonation engine First, the detonation velocities D exp have been
(PDE). With this intent, a model molecule, easier and measured and compared to the theoretical val-
safer to transport than a gas (such as ethylene, which ues D CJ , predicted by the Chapman–Jouguet
is well designed to supply this new kind of propulsion, model [8,9]. These speeds Dexp ∈ [1800 m s−1 ,
according to Desbordes et al. [5]) has been investi- 2400 m s−1 ] ≈ [6500 km h−1 , 8650 km h−1 ] ≈
gated. The high-temperature combustion and the det- [4050 mph, 5350 mph] correspond to Mach num-
onation of gaseous premixed n-heptane/oxygen mix- bers between 5.9 and 9.4 (the sound speed ranges
tures have been experimentally and numerically char- from 305 m s−1 ≈ 1100 km h−1 ≈ 680 mph at
acterized, taking into account the four following as- Φref ≈ 0.4, down to 255 m s−1 ≈ 920 km h−1 ≈
pects. 570 mph at Φref ≈ 2.5).
Second, the cellular and helicoidal detonation
(1) To better understand the oxidation process, ig- structures have been analyzed: typical sizes λ ∈
nition delays τexp ∈ [37 µs, 680 µs] have been [3.0 mm, π · 76 ≈ 240 mm] ≈ [0.12 , 9.4 ] and
measured behind reflected shock waves in a standard deviations λ ≈ 0.3 · λ, have been mea-
shock tube (test section: inner diameter close sured.
to 76 mm ≈ 3.0 and length about 4500 mm ≈ (4) The sensitivity to detonation has been kineti-
177 ) [1], under the following experimental con- cally interpreted [2–4] by applying the theoret-
ditions: ical “0D” model further designed by ZND (Zel’-
• high temperatures T5 ∈ [1290 K, 1600 K] ≈ dovitch, von Neumann, and Döring [10–12]).
[1015 ◦ C, 1325 ◦ C] ≈ [1860 ◦ F, 2420 ◦ F]; The ignition delays τ sim required by this calcu-
• pressures P5 ∈ [250 kPa, 720 kPa] = [2.5 bar, lation have been simulated:
7.2 bar] ≈ [36 psi, 104 psi]; • using S ENKIN with the modified mechanism;
• equivalence ratios Φ ∈ [0.5, 1.5]; • considering the von Neumann parameters T vN
• high dilution in argon: XAr ∈ [0.88, 0.99]. and P vN , i.e., the average of the extremely het-
(2) Several oxidation mechanisms have been tested erogeneous thermodynamical states behind the
to simulate these delays τ exp using S ENKIN shock front.
(C HEMKIN III package) [1]. In particular, the
Curran et al. detailed model [6,7], composed of 1.2. Interest in using Arrhenius-type correlations for
565 species and 2538 reactions, has shown good high-temperature kinetic studies
agreement with experiments. To decrease the dis-
crepancy (typically by a factor 2 down to 1.25), As underlined by Davidson and Hanson [13], de-
five constant rates have been adjusted within their veloping correlations for experimental ignition delays
validity range. The resulting model, designed as is very useful, because:
the “modified mechanism,” is considered for this
investigation. (1) Some modelers require them for their engine pro-
(3) Self-sustained detonations have been experimen- grams or reactive flow codes, rather than using
tally studied in the same shock tube [2–4], using full or reduced kinetic models to describe the
n-heptane/oxygen mixtures: chemical processes.
• initially at room temperature T1 ≈ 295 K ≈ (2) These relations can be helpful for experimental
20 ◦ C ≈ 70 ◦ F; work, by reducing the number of data needed to
• at initial pressure P1 ∈ [200 Pa, 14 kPa] = fully examine the ignition behavior of a particu-
[2.0 mbar, 140 mbar] ≈ [0.03 psi, 2.0 psi]; lar fuel.
• at equivalence ratios Φ ∈ [0.4, 2.5]; (3) These expressions allow comparison of ignition
• diluted or not in argon: XAr ∈ [0, 0.7]. delays, from studies that might have been con-
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 383

ducted with various reaction conditions and term strong flow turbulence behind the shock front, there
definitions. is nearly no time for movement of active species in-
side the gas, neither heat losses to the walls. However,
In addition, some kinetic properties of the det- extrapolation of such a relationship is needed to esti-
onation can be understood, analyzing the induction mate ignition delays in detonation waves.
delays τ sim , as well as the activation temperatures These simplifications explain why Auffret et al.
θ a, sim deduced from them. Of particular interest is [17], for instance, employed this formalism to deter-
the sensitivity of these parameters to: mine the induction distances i = τi · uvN , typical of
the detonations of C2 H2 /O2 mixtures. This calcula-
• the mixture composition: the equivalence ratio Φ tion, based on the ZND model, has enabled them:
and the inert fraction X Ar ;
• the average thermodynamic state in the detona- (1) to evaluate the average self-sustained detonation
tion wave: the von Neumann parameters T vN and cell width as λ ≈ 29·i , in agreement with West-
P vN ; brook et al. [18];
• the mean particle velocity uvN , in the referential (2) to predict the sensitivity to detonation, because
related to the shock front. the initiation energy of spherical detonation is
roughly proportional to λ3 (according to Zel’-
Because the number of elementary reactions in- dovitch et al. [19]) and thus to τi3 .
creases considerably with the length of the alkyl
chain, the combustion mechanism quickly becomes 1.3. Place of the proposed formalism among existing
uneasy to handle for: reduction processes

• complex fuels such as gasoline, kerosene, diesel


In chemical kinetics, the use of polynomials to
(and the equivalent biofuels);
• reactive surrogates (and even isolated “heavy speed up the simulations is not new: Turányi [20,21]
molecules”) representative of these practical has already applied the repromodel. This mathemat-
compositions. ical tool is intended to reproduce the output of the
complex model with high accuracy and high speed.
This is why, despite increasing computing power, The creation of such a model requires extra effort, but
the simulation of the delays τ sim using C HEMKIN or the savings in computer time can be tremendous.
other tools may continue to consume much time in In the case of the kinetic studies dealing with
many cases. Therefore, it can be interesting to use the combustion or atmospheric pollution carried out by
traditional semi-empirical correlation of ignition de- Turányi and by Tomlin et al. [22], the mechanism is
lays: simplified according to a three-step protocol:
 
θa (1) Ordering, for the investigated conditions, of the
τcor = A · [Fuel]α · [O2 ]β · [Inert]γ · exp . (1)
T species considered as:
This simple relation, inspired by the Arrhenius equa- • “redundant” and so likely to be removed from
tion [14], and Semenov’s considerations about the the mechanism during a preliminary simplifi-
thermal autoignition process [15], was first published cation step;
by Lifshitz et al. [16] and widely used since then. Its • “important” and grouped in submodels, there-
success comes from the ability to estimate the em- after mathematically analyzed;
pirical delays τ exp , quickly (from 105 to 108 faster • “necessary” and employed as parameters for
than a standard mechanism) with a convenient accu- the repromodel.
racy (provided that the investigated initial conditions (2) Design of high-order orthonormal polynomials
do not differ much from the ones used to establish the intended to reproduce the characteristic varia-
correlation). tions of the groups composed of the secondary
To properly use this “0D” concept, it matters that variables, in function of the main parameters.
the reactions in the neighborhood do not really change (3) Simplification of these polynomials by the re-
the chemical development, which should only depend moval of the noninfluential terms (selected
on the initial conditions, as if the chemistry scheme among hundreds—or even thousands—of coef-
has been “frozen” since its onset. Such a restriction ficients).
is not problematic in studying the kinetics of the
gas-phase detonation. The involved reactions are so Consequently, the repromodel applied to chemi-
fast that the impact of convection on the local induc- cal kinetics by Turányi and by Tomlin et al. [20–22],
tion process can be reasonably neglected: despite the takes as inputs:
384 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

• the concentrations of the species considered as the objectives here are much more limited than those
necessary; of Turányi and Tomlin et al.: the aim of our behavior
• the global varying parameters such as tempera- model is to estimate the ignition delay from global
ture (or light intensity, in the case of photochem- parameters.
ical reactions). For these high-temperature processes involving a
chemical reactivity likely to develop in the microsec-
As outputs, this polynomial provides the variations of ond range, this duration is indeed meaningful, be-
the concentrations of the necessary species. cause of the actually little time left for mass and heat
Turányi [21] highlights the improvements of his transfer.
method by comparing its performances with those of In addition, ignition delays are experimentally ac-
the original mechanism and of other reduction pro- cessible, thanks to unambiguous measurements based
cesses which are: on:

• “purely mathematical,” like ILDM (intrinsic low- • the pressure registration (in the case of concen-
dimensional manifold); trated mixtures);
• “chemico-mathematic,” like QSSA (quasi-steady • the emission spectroscopy (in most cases), de-
state approximation) and CSP (computational scribing the relative evolution of the concentra-
singular disturbance). tions of excited species (such as OH∗ or CH∗ )
relaxing by fluorescence;
Turányi takes into account a list of criteria, among • the absorption spectroscopy (sometimes), re-
which five will be employed in this article to evalu- lated to the concentration of intermediates at the
ate the performances of the new correlation: ground state, likely to be excited by lasers;
• the chromatography and the spectrometry (very
(1) the human efforts required to prepare the simpli- rarely, because of the little time left for analysis).
fied model;
(2) the applicability field of the model and its im- Finally, the ignition process is quite simple to
portance compared to the domain covered by the model, because of the quasi-simultaneous and large
database; variations of all the kinetic parameters during the ex-
(3) the gain in computing time (typically from 102 to ponential growth of the radical pool generated by the
104 for Turányi’s work); energy release consecutive to the induction stage.
(4) the precision of the results (less than 1% of devi-
ation for the repromodel); 1.4. Objectives and plan of the article
(5) the degree of understanding of the chemical phe-
nomena as described by the model. This article shows how the semi-empirical correla-
tions intended to fit n-heptane ignition delays can be
The latter notion is generally illustrated by two ex- reformulated into polynomials with 3N coefficients,
treme cases: when N ∈ [1, 4] represents the number of adjustable
kinetic parameters likely to be simultaneously chosen
• the behavior model (trivially called black box), among T , P , X Ar , and Φ. These behavior models do
which is easy and quick to handle, but in coun- not aim at correlating the ignition delays measured
terpart provides little (or even no) accessible with a shock tube (as for the majority of the experi-
useful information: for example, a pure “math- mental studies), but are aimed at those calculated us-
ematical model” difficult to safely extrapolate, ing a validated detailed chemical kinetic mechanism,
like an Arrhenius correlation with a few coeffi- in order to provide a database broad and diversified
cients; enough for a discussion.
• the knowledge model, which offers deep in- It will be shown that despite the high computa-
sight, but requires time to be used and know-how tional speed-up aimed at the performances of the tra-
to be properly handled: for example, a well- ditional correlations, the deviations between the pre-
documented detailed mechanism using more or dictions τ sim and their fits τ cor are maintained neg-
less complex formalisms such as those developed ligible compared to the uncertainties of the original
by Lindemann et al. [23] and then extended by model. So, the prediction quality seems preserved by
Troe [24]. this new formalism.
Another advantage of this method is its applica-
This work, related to the combustion behind shock bility to a wide range of temperatures, pressures, di-
waves and in detonation waves, retains the goals of lutions in argon, and equivalence ratios, comparable
the mathematical notion of repromodeling. However, with the validity domain of the detailed mechanism.
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 385

To obtain such results, this article: tude of τ exp ) grows in a quasi-linear way, with T −1 ,
for the high-temperature range, i.e., for temperatures
(1) analyzes the traditional Arrhenius-type correla- above:
tions (Section 2);
(2) confronts these classical expressions with the de- • 850 K ≈ 575 ◦ C ≈ 1070 ◦ F,
tailed model (Section 3); at 500 kPa = 5 bar ≈ 70 psi;
(3) exposes how the new formalism is built (Ap- • 900 K ≈ 625 ◦ C ≈ 1160 ◦ F,
pendix A); at 1.4 MPa = 14 bar ≈ 200 psi;
(4) characterizes its performance (Appendix A): re- • 1000 K ≈ 725 ◦ C ≈ 1340 ◦ F,
production quality and computational speed; at 4.0 MPa = 40 bar ≈ 580 psi.
(5) exhibits the evolutions of the ignition delays τ
(Section 4), then the activation temperatures θ a These critical values, corresponding to the border
(Section 5), predicted by the mechanism for high with the NTC (negative temperature coefficient) do-
temperatures, as illustrations of the potentialities main, are deduced from the investigations carried out
of the proposed formalism. by Chevalier et al. [25].
The interpolation of the traditional semi-empirical
The advanced correlations are intended to enable correlations (expressed as Eqs. (1) and (2)), within
quick modeling of the kinetic properties of the au- their validity domain, gives good results: the ratio
toignition of n-heptane-based mixtures, as well as between the ignition delays τ cor and τ exp generally
their sensitivity to detonation. So, the tool here pro- remains between 0.5 and 2. Despite this uncertainty
posed has been developed for studies dealing with: factor lower than 2, it is often hazardous to extrapo-
late these expressions far from their validity ranges.
• interpretation of the detonation cellular structure This feature explains why the kineticists prefer to use
by means of the “0D” ZND model; mechanisms (detailed, reduced, or even skeletal) that
• analysis of the predictions of a detailed mecha- keep the advantage of knowledge models, with phys-
nism; ical concepts enabling safer extrapolations.
• coupling of chemical kinetics and fluid mechan-
ics in multidimensional simulations. 2.2. Reformulation of the traditional semi-empirical
correlations

2. Analysis of Arrhenius-type formalism in the To highlight the influential physicochemical fac-


case of n-heptane autoignition tors, Eq. (2) can be modified by clarifying the depen-
dence rules between the delay τ cor and:
2.1. Characteristics of traditional semi-empirical
correlations
• the thermodynamic state (T , P );
In the case of the combustion of n-heptane/oxygen/ • the composition, usually expressed by means of:
argon mixtures, the ignition delay τ can be written as – the molar fractions of each component:
the inverse of a fictive global speed law: X C7 H16 , X O2 , and X Ar ;
  – the equivalence ratio Φ and the inert fraction
α β γ θa X Ar .
τcor = A · [C7 H16 ] · [O2 ] · [Ar] · exp , (2)
T
derived from Eq. (1), with By assuming that the ideal gas law is valid for
pressures significantly lower than 10 MPa = 100
• the pre-exponential factor A (here: µs bar ≈ 1450 psi, the concentration [Y] of each species
m3 · (α+β+γ ) mol−(α+β+γ ) ); is easily calculated for a given T , from its partial pres-
• the activation temperature θa = Ea /R (K), with sure P Y (bar). The latter can be expressed in terms of
the activation energy E a (J mol−1 ); the fraction X Y and the total pressure P , according to
• the exponents α, β, and γ (dimensionless), re- PY XY · P
spectively related to the concentrations of the [Y] = = (3)
R·T R·T
fuel n-heptane, oxygen, and the inert argon; co- (R = gas constant), which, combined with Eq. (2),
efficients comparable to the partial orders of a implies
global combustion rate law (up to sign).
A α β γ
τcor = · XC 7 H16
· XO · XAr
This expression provides the evolution of the ex- R (α+β+γ ) 2
   (α+β+γ )
perimental delay τ exp , with a fairly good approxima- θa P
tion. For n-heptane, ln[τexp ] (i.e., the order of magni- × exp · . (4)
T T
386 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

Table 1
Experimental conditions and ignition delays for high-temperature combustions of n-C7 H16 /O2 /Ar mixtures in shock tubes
References X Heptane X Argon Equivalence Pressure Temperature Delay
(mol%) (mol%) ratio Φ (–) P5 (bar) T5 (K) τ (µs)
Vermeer et al. [26] 2.5 70 1.0 1.3 to 4.1 1270 to 1580 13 to 530
Coats et al. [27] 0.09 to 0.7 95 to 98 0.5 to 4.0 1.1 to 2.0 1300 to 2000 30 to 1250
Burcat et al. [28] 0.5 to 3.0 64 to 88.5 0.5 to 2.0 2.0 to 11.8 1135 to 1660 12 to 647
Colket et al. [29] 0.2 95.6 and 97.7 0.5 and 1.0 4.1 to 7.8 1225 to 1430 88 to 844
Horning et al. [30] 0.2 to 1.8 78 to 97.6 0.5 to 2.0 1.1 to 5.7 1330 to 1680 86 to 490
Imbert et al. [1] 0.08 to 1.0 88.0 to 99.0 0.5 to 1.5 2.5 to 7.2 1290 to 1600 37 to 680
Smith et al. [31] 0.4 90.8 to 97.4 0.5 to 2.0 1.0 to 2.0 1150 to 1650 40 to 1000

P MMixture
This expression of τ cor exhibits the “pressure coeffi- ρ= · , (10b)
cient” p, defined as the global order (i.e., the sum of T R
the partial orders of every species): where M Mixture represents the average molar mass of
the mixture.
p = (α + β + γ ). (5)
τ cor can be also linked to Φ, since in the case of 2.3. Analysis of the traditional semi-empirical
{C7 H16 /O2 } mixtures
correlations
XC7 H16 /XO2
Φ= ⇔ (6a)
1/11 2.3.1. Expressions published for n-heptane
XC7 H16 Φ autoignition
= , (6b)
X O2 11 Table 1 lists in the case of n-heptane:

1 − XAr = XReac = XO2 + XC7 H16 • the shock tube ignition delays τ exp , already pub-
 
XC7 H16 lished [1,26–31];
= XO2 · 1 + • the experimental conditions used for these high-
X O2
  temperature investigations.
Φ
= XO 2 · 1 + , (7)
11
Table 2 exposes the semi-empirical correlations de-
and therefore duced from these measurements and reformulated to
 −(α+β) facilitate their analysis and comparison.
A α · 1+ Φ
τcor = · Φ
11α · R (α+β+γ ) 11
γ 2.3.2. Role of the temperature T
XAr
× As demonstrated by Arrhenius, temperature plays
(1 − XAr )−(α+β) a major role in chemical kinetics. Due to the break-
   (α+β+γ ) ing of bonds, during the initiation process, the slope
θa P
× exp · . (8) of the quasi-linear relation ln(τexp ) = f (T −1 ), re-
T T
lated to the induction time, remains generally very
This expression of τ cor exhibits the “implicit coeffi- high. Here, the deviations between the activation tem-
cient related to dilution.” This parameter, used as an peratures θa ∈ [17.8 × 103 K, 35.2 × 103 K] (activa-
exponent for (1 − XAr ), can be defined as the opposite tion energies Ea ∈ [148 kJ mol−1 , 293 kJ mol−1 ] ≈
of the sum of the partial orders linked to the reactants: [35.3 kcal mol−1 , 70.0 kcal mol−1 ]), measured by all
the authors have been explained [1]:
x = −(α + β). (9)
For CFD (computational fluid dynamics), density • by the difference in amplitude (from 200 K (or
ρ can be a more convenient parameter, although it ◦ C), i.e., 360 ◦ F, to 700 K, i.e., 1260 ◦ F), and po-
cannot be directly evaluated through experiments in sition (maximum between 1430 K ≈ 1160 ◦ C ≈
shock tubes. Small adaptations of Eq. (4) (and follow- 2110 ◦ F and 2000 K ≈ 1730 ◦ C ≈ 3140 ◦ F) of
ing expressions directly based on it) are then needed. the investigated temperature ranges;
Practically, it is enough to replace P /T by ρ, using • by the diversity of the dilution importance from
P R less than 70% to more than 99% of the mixture
=ρ· ⇔ (10a) filled with argon.
T MMixture
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 387

Table 2
Expression of the semi-empirical correlations τ cor (µs) (established from the delays τ exp measured for the conditions described
in Table 1) as a function of temperature T (K), pressure P (bar), and the concentrations [Mixture], [C7 H16 ], [O2 ], and [Ar]
(mol m−3 )
References Semi-empirical correlations x p
 
Vermeer et al. [26] τcor = 7.34 × 10−5 · [Mixture]−0.86 · exp 23T340 +0.86 −0.86
 
Coats et al. [27] τcor ∝ [C7 H16 ]+1.42 · exp 35T300 – –
 
Burcat et al. [28] τcor = 2.0 × 10−4 · [C7 H16 ]+0.2 · [O2 ]−1.1 · [Ar]+0.6 · exp 17T765 +0.9 −0.3
 
Colket et al. [29] τcor = 4.26 × 10−4 · [C7 H16 ]+0.4 · [O2 ]−1.2 · [Ar]0 · exp 20T211 +0.8 −0.8
 
Horning et al. [30] τcor = 2.73 × 10−4 · [C7 H16 ]+0.95 · [O2 ]−1.58 · [Ar]0 · exp 22T647 +0.63 −0.63
 
Imbert et al. [1] τcor = 1.21 × 10−5 · [C7 H16 ]+0.92 · [O2 ]−1.41 · [Ar]+0.09 · exp 26T922 +0.49 −0.40
 
Smith et al. [31] τcor = 10(−3.30±0.04) · (1.013 · P )(−0.468±0.043) · exp 20 650±139
T – −0.468
Evaluation of the “implicit coefficient of dilution” x = −(α + β) and of the “pressure coefficient” p = (α + β + γ ), using the
“partial orders” α, β, and γ (dimensionless), respectively related to [C7 H16 ], [O2 ], and [Ar].

⎧  k1 ·k2 
2.3.3. Role of the pressure P ⎪
⎪ · [AB] (molecularity 1),

⎨ −1
k
Generally speaking, the reactivity grows with at high pressure,

pressure, which implies that the sensitivity to pres- ⎪

sure factor p = (α + β + γ ) remains negative (here: ⎩ (k1 · [M]) · [AB] (molecularity 2),

p ∈ [−0.83, −0.3]). at low pressure.
Several kinetic approaches enable the qualitative (11)
understanding of this power-type dependence, and As concentrations of the main active species dur-
notably: ing the autoignition process are often low, the tran-
sition pressures, although evolving with temperature,
• the “fall-off” model developed by Lindemann et
are generally worth about several bars. Because these
al. [23], which explains the change in the molec-
pressures are much higher than the ones investigated
ularity (“number of reactants”) of the elementary
reactions with pressure; here, the apparent molecularities of the elementary re-
• the RRKM theory [32], which provides simple actions involved during induction often exceed unity.
estimates of the unimolecular reaction rates, from The exponent p, bounded to the “global” reaction,
a few characteristics of the potential energy sur- results from the overall combination of each molec-
face. ularity, weighted by the importance of the related el-
ementary reaction. Consequently, its determination is
In accordance with these models: complex. For example, it may depend on the compe-
tition between two elementary reactions, one second
(1) The elementary reaction AB (+M) → A + B order and another third order, which are respectively
(+M) proceeds according to the following global especially active at low and high pressures.
scheme, composed of three steps: Actually, the pseudo-order of the “global” reaction
• AB + M → AB∗ + M is given by (1 − p): for example, p = 0 would corre-
(k1 ): activation of the reactant AB by the col- spond to the case of the strictly unimolecular reaction,
lision partner M; for which the ignition delay does not depend on pres-
• AB∗ + M → AB + M sure. Here (1 − p) is intermediary between 1 and 2.
(k−1 ): deactivation (also by collision) of the
This value implies that the reactions involved by the
reaction intermediate AB∗ , vibrationally ex-
autoignition process (mainly the initiation stage of the
cited;
chain reaction, which is combustion) are not really
• AB∗ → A + B
(k2 ): dissociation of AB∗ into products A unimolecular due to:
and B.
(2) The speed of this “so-called elementary reac- (1) the importance of the collision partner, as de-
tion,” assuming QSSA for AB∗ , is scribed by the Lindemann “fall-off” model;
  (2) the presence of reactions involving two or more
d[AB] k1 · k2 · [M] reactants, that is, reactions other than decompo-
v=− = · [AB]
dt k−1 · [M] + k2 sition, β scission, isomerization, etc.
388 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

2.3.4. Role of the inert or reactive fraction XAr and Table 3


XReac Experimental validation range associated with the Curran et
Concerning the dilution impact, the correlations al. original mechanism [6,7] and conditions for simulation
(Table 2) confirm the intuition: τ cor increases with using the version modified by Imbert et al. [1]
X Ar , all other parameters being equal: Parameter Validation Conditions
range of use
γ  0 and x = −(α + β) > 0 ⇒ Pressure P (bar) 1 to 42 0.1 to 10
γ Temperature T (K) 550 to 1700 1300 to 1900
XAr
τcor ∝ strictly grows with XAr . Equivalence ratio Φ (–) 0.3 to 1.5 0.25 to 4.0
(1 − XAr )x Argon fraction X Ar (–) 0.70 to 0.99 0 to 0.70
This dependence type shows that argon addition in- Reactive fraction X Reac (–) 0.01 to 0.30 0.30 to 1.00
creases τ cor :
2.4. “Linearization” of Arrhenius-type expressions
• explicitly, through the coefficient γ of [Ar]: tak-
ing into account the singular effect of argon, as To continue the calculation development, the
a special collision partner to deactivate the in- semi-empirical correlation (Eq. (8)) is linearized by
termediate reactants (to be noted that γ is often applying a logarithm transform:
neglected and imposed on zero, as seen in Ta-
ble 2); lg(τcor ) = Lτref + f (Φ) + γ · lg(XAr )
• implicitly, with the coefficients α and β related θa
to the concentrations [C7 H16 ] and [O2 ]: mani- − x · lg(1 − XAr ) +
T
festation, through the so-called “implicit dilution − p · lg(T ) + p · lg(P ) (12)
coefficient” x = −(α + β), of the rarefaction of
these two reactants. with:

A
• Lτref = lg( α (α+β+γ
2.3.5. Role of the proportion between reactants 11 ·R ) ), later directly evalu-
(equivalence ratio Φ) ated, and assumed as a constant;
• f (Φ) = α · lg(Φ) + x · lg(1 + 11 Φ ) complex de-
The correlations (Table 2) show a dependence of
τ cor : pendence, reformulated thereafter to keep the lin-
earity;
• slightly increasing with [C7 H16 ], since α  0.2; • γ = 0: assumption often used (Table 2), and here
• definitely decreasing with [O2 ], since β  −1.1; intended to avoid a discontinuity for the undi-
• increasing with Φ, and varying in luted mixtures (practically for very low values of
XAr ∈ [0, 0.7]).
 
Φ −(α+β)
Φα · 1 + (deduced from Eq. (8))
11
3. Confrontation between Arrhenius formalism
with: −(α + β) = x  0.49. and mechanism predictions

It is worth noting that n-heptane—though reactive 3.1. General information concerning analysis of
—inhibits its own ignition (α > 0), contrary to oxy- simulated delays τsim
gen, which definitely promotes it (β < 1). This fea-
ture is usual with alkanes, but not general for all the The program S ENKIN (C HEMKIN III package) has
fuels (for example, monomethylhydrazine CH3 N2 H3 , enabled the prediction of ignition delays τ sim , defined
studied by Catoire [33]), depending on their chemical as the time interval between the onset of the shock
natures. For high-temperature combustion, this be- wave and the maximum of dT /dt curve: the temper-
havior especially results from a competition between ature rise inflection generally occurs simultaneously
fuel and oxygen, for reactions with ◦ H atoms, accord- with the emission peak related to the excited radical
ing to: ◦ OH∗ .
These delays τ sim have been evaluated using the
• RH + H◦  R◦ + H2 : metathesis generating a Curran et al. mechanism [6,7], somewhat modified
radical R◦ heavier than ◦ H, thus less reactive; by Imbert et al. [1]. The conditions employed here to
• O2 + H◦  ◦ OH + O: chain branching, leading model the ignition have slightly surpassed the experi-
to the “exponential” growth of the number of re- mental validation range (Table 3), as is often the case
action intermediates, and causing the very quick when a mechanism is used, especially with regard to
energy release during an explosion. the high temperatures.
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 389

Basically for the calculation, the combustion has


been assumed to develop in a constant volume, with-
out heat losses, since:

• the shock tube is a closed reactor;


• the ignition delays are too short, to allow:
– the pressure waves to relax far from the induc-
tion zone;
– the gas core to be cooled by an exchange with
the boundary layers bordering the wall.

With a Pentium III running at 500 MHz, about 15 min


of calculations were required for each ignition delay.
Figs. 1–4 illustrate the evolution of the delays with
the parameters T , P , Φ, and XReac = 1 − XAr . The
performances of the various relationships presented
in Table 1 will be quantified by the standard devi-
ation τ cor , between the simulated ignition delays Fig. 1. Evolution of the ignition delays (P for the reference
τ sim , used as references, and their fits τ cor (defini- simulated value τ sim ; dashed line for the correlation τ cor )
tion expressed by Eqs. (A.17) and (A.18) in Appendix with the temperature T and its inverse T −1 (normal–log
A.4). scale), around T ref = 1700 K. Arrhenius diagram estab-
All the parameters presented in this section (the lished for the mixture n-C7 H16 /O2 (Φref = 1) at P ref = 100
kPa. Reference points obtained with the modified mecha-
activation energies θ a, sim , the pressure coefficients
nism (Imbert et al. [1]).
p, the implicit dilution coefficients x, and the partial
pseudo-orders α related to n-heptane) will provide a
3.2. Evolution of τsim with T
frozen image of the predictions of the detailed mecha-
nism for a particular condition (here, the one arbitrar-
Clearly, the temperature remains the parameter
ily chosen as a reference).
with the most influence on the ignition delay: as illus-
It is difficult to deeply interpret the differences and
trated in Fig. 1, τ sim shortens by about a factor 140
the similarities between the simulated and the semi- (from 280 µs down to 2.0 µs) when T increases by
empirical values. This work, close to the comparison 45% (from 1300 K up to 1900 K).
between various Arrhenius-type correlations related The dashed line (inspired by the original Arrhe-
to the same fuel, is beyond the scope of this study. nius rate law) correctly fits τ sim , although τcor > τref
At this stage of the kinetic study, it is only possible in the center of the graphic (line above the triangles)
to list, as did Davidson and Hanson [13], the possible and τcor < τref at its ends (line below the triangles):
origins of the disagreements between semi-empirical
 
τcor T
correlations: = exp −11.6 · 1 − ref
6.13 T

 
• the use for conditions outside the validity do- 3 1 1
= exp 19.7 × 10 · −
main; T Tref
• the differences in the definition of the ignition de- ⇒ τcor = 5.0%. (13)
lays (especially the choice of the time markers);
• the errors in the experimental methods (partic- The mathematical use of a second-order polynomial
in T −1 enables better reproduction of the slight con-
ularly the limitations inherent to the shock tube
vexity, source of the deviation τ cor between the
and to the diagnostics).
simulated delays τ sim and their fit τ cor :
   
In addition, it must be kept in mind that the discrep- τcor 1 1
ln = 18.9 × 103 · −
ancies with the measured ignition delays τ exp are im- 6.01 T Tref
portant:  
1 1 2
+ 7.20 × 106 · −
• at least 25%, for the duration τ sim , simulated by T Tref
the modified detailed mechanism; ⇒ τcor = 0.7%. (14)
• typically a factor 2, for the time τ cor , calculated It is also possible to use the modified Arrhenius-type
using the semi-empirical correlations. law by putting an exponent on T (here, the value is
390 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

directly deduced from the best fit):

 
 
τcor T 11.5 1 1
= · exp 38.3 × 103 · −
6.01 Tref T Tref
⇒ τcor = 0.5%. (15)

The formalism of Eq. (15) is inspired by expression of


the elementary rate constants. On one hand, for a data-
processing complexity comparable with the second-
order polynomial fit (Eq. (14)) in T −1 , this expres-
sion reduces the deviation τ cor . On the other hand,
correlation (15) masks θ a, cor , which now requires the
intermediate calculation of at least two τ cor delays to
be evaluated. Despite this disadvantage, this type of
formalism will be selected to develop the new corre-
lations.
The apparent activation temperatures θ a, sim , es-
tablished by the linear regression ln(τcor ) = f (T −1 ) Fig. 2. Evolution of the ignition delays (P for the reference
around T ref , remain almost the same for Eqs. (13)– simulated value τ sim ; dashed line for the correlation τ cor )
(15), respectively 19.7 × 103 K, 18.9 × 103 K, and with the pressure P (log–log scale), around P ref = 100 kPa,
18.8 × 103 K, corresponding to activation energies for the mixture n-C7 H16 /O2 (Φref = 1) at T ref = 1700 K.
E a, sim about 160 kJ mol−1 or 38 kcal mol−1 . This Reference points obtained with the modified mechanism
property shows that the slight convexity of ln(τsim ) = (Imbert et al. [1]).
f (T −1 ), although not easily visible in Fig. 1, weakly
   
influences the simulated apparent activation temper- τcor P
atures θ a, sim . Moreover, they remain close to the lg = −0.763 · lg
5.94 Pref
semi-empirical values, deduced from n-heptane cor-
  2
relations (Table 2): P
+ 0.052 · lg
Pref
• θa, cor ≈ 17.8 × 103 K, for Burcat et al. [28]; ⇒ τcor = 0.1%. (17)
• θa, cor ≈ 20.2 × 103 K, for Colket and Spadaccini
[29]; The coefficients psim ≈ −0.762 exhibited by Eqs.
• θa, cor ≈ 20.7 × 103 K, for Smith et al. [31]. (16) and (17) remain very close to the sensitivity fac-
tor to the pressure deduced from the semi-empirical
3.3. Evolution of τsim with P correlations (Table 2): pcor ≈ −0.8 from Colket and
Spadaccini [29], for example.
Pressure has a smaller influence on the ignition
delays than temperature: whereas the range covered 3.4. Evolution of τsim with XAr and XReac
by P extends much more (in logarithmic scale) than
the one related to T , the impact on τ sim remains of The mixture dilution has also a secondary influ-
the same order of magnitude in both cases: as illus- ence on the ignition delays: as illustrated in Fig. 3,
trated in Fig. 2, τ sim only shortens by about a factor τ sim does not increase more than by a factor 10
30 (from 38 µs down to 1.2 µs) when P increases by (from 5.9 µs up to 55 µs), when the reactive fraction
a factor 100 (from 100 kPa up to 10 MPa). XReac = XC7 H16 + XO2 = 1 − XAr decreases by a
Due to the quasi-linear relationship between P factor 100 (from 1 down to 0.01).
and τ sim , for a log–log scale, the dependence seems The quasi-linear relationship between X Reac and
to be of power type τ sim , for a log–log scale, leads to:

     
τcor P −0.762 τcor XReac −0.456 1 − XAr −0.456
= ⇒ τcor = 6.8%. (16) = =
6.25 Pref 6.15 1 1 − XAr,ref
⇒ τcor = 7.2%. (18)
The use of a second-order polynomial for the consid-
ered scale here enables a better reproduction of the The use of a second-order polynomial reduces the dis-
convexity with: crepancy:
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 391

plained by Davidson and Hanson [13], as well as by


Brun [34], the use of a minimum amount of an inert
“bath gas” enables the experimentators:

(1) to significantly lower the reaction mass enthalpy


and thus to limit the risks of an accidental deto-
nation onset;
(2) to absorb the energy variations behind the shock
waves (effect of “heat sink”) and thus:
• to homogenize the thermodynamic state;
• to attenuate the impact of the stochastic distur-
bances, especially in the medium shocked by
the reflected wave, which is very sensitive to
the boundary layers’ perturbations left by the
incident wave passage;
(3) to study the global kinetics for high tempera-
ture, while keeping τexp ∈ [10 µs, 1 ms] observ-
able and realistic to study the detonation: the dif-
Fig. 3. Evolution of the ignition delays (P for the reference ference in the induction delay related to a pure
simulated value τ sim ; dashed line for the correlation τ cor ) mixture is approximately only an order of mag-
with the fraction of the reactant X Reac (log–log scale) and
nitude;
of the dilutant X Ar , for the mixture n-C7 H16 /O2 /Ar (Φref =
1) at T ref = 1700 K and P ref = 100 kPa. Reference points
obtained with the modified mechanism (Imbert et al. [1]). and in the special case of the use of a monoatomic
“bath gas” like argon (instead of the diatomic nitro-
    gen, employed to simulate air):
τcor 1 − XAr
lg = −0.527 · lg
5.96 1

  (4) to carry out this temperature rise by still generat-
1 − XAr 2 ing a moderate shock (Mach number Ms ≈ 2.5):
− 0.044 · lg
1 the energy provided by the shock is then in ma-
⇒ τcor = 1.2%. (19) jority distributed to the degree of freedom in
translation, and only a small part is lost in the lev-
These results are good with regard to the impor- els of vibration and rotation present in the poly-
tance of the dilution range (2 orders of magnitude for atomic reactants;
XReac ∈ [0.01, 1]). Consequently, the presence of the (5) to limit the influence of the vibrational relaxation
γ
multiplicative term XAr in Eqs. (4) and (8) (which in the carrier gas and so to decrease the duration
causes continuity troubles through the logarithmic ex- required to reach the thermodynamic equilibrium
pression for XAr = 0) does not seem necessary to behind the shock waves;
correct the inflection of ln(τsim ) = f [ln(XAr )]. So, (6) to minimize the bifurcation of the reflected shock
“the partial order” γ , which exhibits the explicit influ- wave during its passage inside the boundary lay-
ence of argon (Section 2.3.4), can be set to zero with- ers (generated near the wall by the previous pas-
out problem. As illustrated in Table 2, this is done in sage of the incidental shock), consequence of the
the majority of the semi-empirical correlations (γ = 0 fifth item.
γ
implies XAr = 1, whatever the values of XAr ∈ [0, 1]
and XReac = 1 − XAr ). On the other hand, these experiments are carried
Consequently, only the implicit kinetic effect (re- out under conditions relatively far from those of prac-
lated to the decrease of the reactant concentrations) tical combustion devices (engines or turbines):
plays a role. The absolute values of the exponents
x sim , −0.456 and −0.527 (slopes of τcor = f (XReac ) • air contains only 79% of inert rather than 98%
for XReac = 0, according to a log–log scale), can (or even more);
be compared to the implicit dilution coefficients de- • for the same inert fraction and energy release,
duced from the semi-empirical correlations (Table 2): the high isentropic coefficient of argon (1.67 for
xcor = 0.49 for Imbert et al. [1], for instance. this monoatomic gas, greater than the diatomic
In addition, the minor impact on kinetics of the di- nitrogen one: from 1.4 at 300 K down to 1.3 at
lution (in comparison to the temperature one), a pos- 2000 K) results in large variations of the mixture
teriori validates the measurement protocol of the temperature during the induction process, and so
shock tube ignition delays τ exp (Table 1). As ex- possibly in a change of the ignition delay τ exp .
392 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

According to the n-heptane detailed oxidation mech-


anism (Fig. 4), the evolution of τ sim with Φ can be
described by three branches, thereafter qualified by:

• very lean: Φ  0.15;


• lean, up to stoichiometric: Φ ∈ [0.15, 1.0];
• rich: Φ > 1.0.

In agreement with the log–log scale, the associated


correlations are illustrated by:

• short dashed lines for the extreme zones (very


lean and rich mixtures);
• a long dashed parabola for the central transition
zone (fairly lean medium and stoichiometric).

These functions are extrapolated beyond their validity


Fig. 4. Evolution of the ignition delays (P for the reference domains to underline the discrepancy (reference τ sim
simulated value τ sim ; dashed line for the correlation τ cor ) (P); correlation τ cor (– –)), which could generate er-
with the equivalence ratio Φ (log–log scale) for the mixture rors in the case of a bad division of the equivalence
n-C7 H16 /O2 at T ref = 1700 K and P ref = 100 kPa. Refer-
ratio range.
ence points obtained with the modified mechanism (Imbert
et al. [1]). Legend for the correlations extrapolated beyond
To facilitate comparison between the semi-empir-
their validity range: long dashed line, decreasing with Φ, for ical correlations and the mechanism predictions, only
the very lean mixture; short dashed parabola, for the fairly the evolution of Φ α will be considered. Indeed, the
lean mixture; long dashed line, increasing with Φ, for the fractional part of Eq. (20) can be neglected, because:
rich mixture.
• it always remains close to unity;
Consequently, disagreements between predictions of • it slowly and weakly varies for rich mixtures:
mechanisms validated with mixtures very diluted in no more than 15% for Φ ∈ [1, 4] by supposing
argon and direct measurements with air are often ob-
α + β ≈ −xsim ≈ −0.5;
served.
• it remains quasi-constant for lean mixtures: ap-
To conclude this issue, if the model correctly re-
proximately (12/11)−0.5 ≈ 0.96 for Φ 1
produces the evolution of the ignition delays with
11.
temperature, it can have difficulties making the same
with less influential kinetic parameters (such as di-
lution), due to the choice of the experimental condi- 3.5.2. Rich mixtures (Φ > 1.0)
tions. For rich mixtures, n-heptane inhibiting effect pre-
vails, due to the important rerouting of ◦ H atoms,
3.5. Evolution of τsim with Φ from chain branching (an essential reaction for the
3.5.1. General high temperature combustion). Due to this feature,
The equivalence ratio has a secondary influence on τ cor increases with Φ:
the ignition delays: as illustrated in Fig. 4, τ sim does
not vary by more than a factor 10 (at most between  
τcor Φ 1.08
2.6 and 26 µs), when Φ evolves by a factor 80 inside = ⇒ τcor = 1.5%. (21)
5.93 Φref
[0.05, 4.0] range intended to cover the flammability
domain of n-C7 H16 /O2 .
This exponent αsim = 1.08 remains close to the par-
Certainly due to the complexity of the reaction
tial orders α cor , deduced from the semi-empirical cor-
pathway, the simulated evolution cannot be repro-
duced in this whole area by the semi-empirical cor- relations (Table 2):
relations: the part of Eq. (8), restricted to the depen-
dence versus Φ varies as • αcor ≈ 0.95, for Horning et al. [30];
 
    α+β • αcor ≈ 0.92, for Imbert et al. [1].
τcor Φ α Φ Φ
= · 1 + ref 1+
τref Φref 11 11 3.5.3. Very lean mixtures (Φ  0.15)

α+β For very lean mixtures, the small presence of hep-
12
= Φα · . (20) tane:
11 + Φ
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 393

• rarefies the collisions between oxidizing and • a parabola (traditional mean in mathematics):
reducing species (unfavorable kinetic contribu-  
τcor
tion); lg = 1.12 · lg(Φ) + 0.92 · lg(Φ)2
• limits the rerouting of ◦ H atoms (favorable fac- 5.86
tor). ⇒ τcor = 1.9%; (23)
• a higher order polynomial (for better agreement):
In this extreme case, due to the prevalence of the first  
effect, τ cor decreases with Φ: τcor
lg = 1.22 · lg(Φ) + 1.19 · lg(Φ)2
5.93
  + 0.20 · lg(Φ)3
τcor Φ −0.42
= ⇒ τcor = 0.7%. (22)
1.29 Φref ⇒ τcor = 0.9%. (24)

Here, n-heptane behaves like a promoter of its own The use of a complex polynomial requires more cal-
ignition and no longer like an inhibitor. culations:
Incidentally, it is difficult to give much credit to
the prediction of delays τ sim reported for Φ  0.15, • to evaluate the correlation;
due to: • to determine a database related to reference con-
ditions.
• the use of the mechanism in a range where it
has not been validated by experiments (even if This constraint is not justified to study detonation ki-
one big advantage of this knowledge model is the netics, since the range to be reproduced is limited
possibility to extrapolate quite safely); to reactive explosive mixtures, and this is practically
• the great difficulty of defining ignition delays, us- quite close to stoichiometry.
ing S ENKIN simulations: A smart mathematical solution thus consists of:
– progressive and slow increase in temperature
and pressure without marked inflection; • restricting the interpolation range only to the use-
– complex exploring of the time profile of the ful zone: Φ ∈ [0.25, 1.0], for example;
◦ OH radical concentration, which is the only • simply using a parabola (log–log scale):
means to define ignition delays for those very  
τcor
lean conditions. lg = 1.17 · lg(Φ) + 0.99 · lg(Φ)2
5.91
⇒ τcor = 0.6%. (25)
3.5.4. Lean and stoichiometric mixtures
(Φ ∈ [0.15, 1.0])
The evolution of τ sim is correctly reproduced under
For the intermediate range of Φ, due to the previ-
conditions of forced quasi-tangency between the cen-
ously mentioned evolutions of τsim = f (Φ), the re-
tral parabola and the extreme line related to the rich
actions become the fastest, any other parameters (T , mixture. This connection condition between lean and
P , and X Ar ) are fixed. For the reference conditions rich mixtures will be respected to establish the corre-
(Tref = 1700 K, Pref = 100 kPa), the minimum delay lations reported in Appendix A (Tables 4–7).
τ sim, min is obtained for the optimal equivalence ratio
Φopt,τ ≈ 0.25. These lean conditions (actually quite 3.5.5. Conclusion about the impact of Φ on τsim
far from stoichiometry) correspond to a compromise The power-type law in Φ α for τ cor seems well
where n-heptane is: adapted, where one of the reactants is clearly in ex-
cess:
• abundant enough, so that the reactive species can
frequently collide during the propagation stages; • for rich mixtures, the exponent 1.08 borders the
• in low enough proportion that a large number of partial order α linked to [C7 H16 ], deduced from
◦ H atoms remain, despite their capture by alkyl
the semi-empirical correlations;
radicals issued from the alkane, during metathe- • for very lean mixtures (Φ  0.15 11), the ex-
sis reactions. ponent −0.42 should be directly related to α (ac-
cording to Eq. (20)), but in fact takes an opposite
Since this zone corresponds to the local minimum sign.
of τsim = f (Φ), it can be correlated in good approx-
imation by the “flattened” vicinity of the bottom re- Since the absolute value of −0.42 is small compared
lated to: to 1.08, the negative impact (caused by the lack of
394 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

Table 4
Coefficients related to Eq. (A.19) reproducing the delays τ sim , with a deviation lower than 1.5%
τ ref, 1 (µs) a1 b1 10 · c1 d1 e1 100 · f1 g1 h1
5.9492 −9.8017 11.556 −7.7446 1.4355 −4.1649 5.1452 1.3880 −2.8054
Simulation related to the mixture (n-C7 H16 /O2 , X Ar = 0 and Φ = 1) initially at T ∈ [1500 K, 1900 K] and P ∈
[10 kPa, 1 MPa].

Table 5
Coefficients related to Eq. (A.20) reproducing the delays τ sim , with a deviation lower than 1.5%
τ ref, 2 (µs) a2 b2 10 · c2 d2 e2 100 · f2 g2 h2
5.9492 −9.7926 11.524 −7.6260 1.4388 −4.1716 5.1511 1.3880 −2.8022

10 · i2 j2 k2 1000 · l2 m2 n2 1000 · o2 p2 q2
−5.9470 3.8080 −8.9598 4.3591 −1.2543 3.2734 9.3907 3.2218 −7.2594

10 · r2 s2 t2 100 · u2 10 · v2 w2 100 · x2 y2 z2
−1.4835 −3.5744 9.6339 4.4238 6.9279 −1.9606 3.8142 6.8344 −15.878
Simulation related to the diluted mixture (n-C7 H16 /O2 /Ar, Φ = 1 and XAr ∈ [0, 0.7]) initially at T ∈ [1500 K, 1900 K] and
P ∈ [10 kPa, 1 MPa].

Table 6
Coefficients related to Eq. (A.21) reproducing the delays τ sim , with a deviation lower than 2.5%
τ ref, 3 (µs) a3 b3 10 · c3 d3 e3 100 · f3 g3 h3
5.9639 −9.5896 11.028 −7.6208 1.5059 −4.3281 5.0315 1.1765 −2.2888

i3 j3 k3 10 · l3 m3 n3 100 · o3 p3 q3
1.0947 12.368 28.016 −2.1004 −6.1010 15.874 −6.9263 3.7162 −8.9187
Simulation related to the rich mixture (n-C7 H16 /O2 , Φ ∈ [1.0, 4.0] and XAr = 0) initially at T ∈ [1500 K, 1900 K] and
P ∈ [10 kPa, 1 MPa].

Table 7
Coefficients related to Eq. (A.22) reproducing the delays τ sim , with a deviation lower than 3.5%
τ ref, 4 (µs) a4 b4 10 · c4 d4 e4 100 · f4 g4 h4
5.9594 −9.6144 11.092 −7.6220 1.4998 −4.3167 5.0400 1.1954 −2.3350
i4 j4 k4 10 · l4 m4 n4 10 · o4 p4 q4
1.1509 21.600 −53.524 −1.2920 −1.3344 6.2639 −1.0040 −3.5367 8.8039

10 · r4 s4 t4 10 · u4 v4 w4 100 · x4 y4 z4
9.4960 29.676 −65.913 −1.6400 −12.332 31.079 −7.0100 −5.5594 11.871
Simulation related to the lean mixture (n-C7 H16 /O2 , Φ ∈ [0.5, 1.0] and XAr = 0) initially at T ∈ [1500 K, 1900 K] and
P ∈ [10 kPa, 1 MPa].

heptane for very lean mixtures) seems less marked • have little interest for practical kinetic studies re-
than the positive one (resulting from its excess for rich lated to the combustion in engines or turbines (in
mixtures). both the deflagration and the detonation modes).
These temporary conclusions, dealing with very
lean mixtures, have only small practical meaning, 3.6. Temporary conclusion and formulation of
since these conditions: significant reduced variables
This paragraph describes the relevance of trans-
• remain little explored during measurements of posing the formalism deduced from semi-empirical
delays τ exp , due to the proximity of the lower correlations to detailed mechanism predictions. Ex-
explosion limit (for n-heptane, at ambient tem- cept for the dependence (τsim | Φ) related to lean
perature: XC7 H16 ≈ 1.05% and Φ ≈ 0.05); mixtures, the agreement (traditional semi-empirical
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 395

correlation | detailed mechanism) is good with regard τ sim and of the activation temperatures θ a, sim for n-
to: heptane combustion at high temperature is based on
the use of polynomial correlations such as
• the general evolutions of τsim = f (T , P , XAr ,
Φ);  
2
2
2
τcor
• the numerical values of the deduced coefficients: lg = LΦ k · LXj · LPi
τref
θ a, sim , p sim , x sim , and α sim . k=0 j =0 i=0
× (ai,j,k + bi,j,k · IT + ci,j,k · LT).
By considering a wide range of thermodynamic (31)
states and mixture compositions,
These “linear” relations try to conciliate:
• T ∈ [1500 K, 1900 K] ≈ [1225 ◦ C, 1625 ◦ C] ≈
[2240 ◦ F, 2960 ◦ F], centered in Tref = 1700 K ≈ • the prediction precision and the flexibility of
1425 ◦ C ≈ 2600 ◦ F, the detailed mechanisms (knowledge models) for
• P ∈ [10 kPa, 1 MPa] = [100 mbar, 10 bar] ≈ various thermodynamic conditions and mixture
[1.45 psi, 145 psi], centered in Pref = 100 kPa = compositions;
1 bar ≈ 14.5 psi, • the simplicity and the calculation speed of either
• Φ ∈ ([0.4, 1.0] ∪ [1.0, 4.0]), centered in Φref = the semi-empirical correlations or the global re-
1.0, action (both behavior models).
• XAr ∈ [0, 0.7], “centred” in X Ar, ref = 0,
Appendix A shows how:
it thus seems possible to correlate the linearized form
lg(τcor ) = f (T , P , Φ, XAr ) correctly with a mini- (1) to arrive at this polynomial fit through a mathe-
mum set of coefficients. matical development of increasing complexity;
The impact analysis of the influential parameters (2) to accelerate the numerical evaluation of τ cor
suggests to use as reduced variables: (useful mean in the case of an intensive use, as re-
quired in simulation of processes related to com-
T
• IT = 1 − ref (26) bustion behind shocks);
T (3) to efficiently and quickly prepare a kinetic data-
and its correction base;
 
T (4) to simply and rapidly establish the polynomial
LT = lg (27)
Tref coefficients (Moore and Penrose’s analytical res-
to reproduce a modified Arrhenius-type law for olution [35], here explicitly used for four investi-
the temperature (major kinetic factor); gation ranges);
  (5) to evaluate the reproduction performances of
P
• LP = lg (28) τ sim by τ cor :
Pref
• the deviation is lower than 3.5% and often
and its correction LP2 for the pressure; close to 0.5%;
   
XReac 1 − XAr • the calculation is about 106 faster than the
• LX = lg = lg (29)
XReac,ref 1 − XAr,ref step-by-step resolution using S ENKIN, and
performance is comparable with the traditional
and its correction LX2 for the fraction of either semi-empirical correlation (this computational
dilutant X Ar or reactant X Reac ;
  speed-up does not depend on the computer
Φ
• LΦ = lg (30) power, but on the complexity both of the cor-
Φref relation and of the mechanism).
and its correction LΦ 2 for the equivalence ratio.
The slight differences (correlation τcor | reference
τsim ) are satisfactory, for two reasons:
4. Evolution of the ignition delays τ sim simulated
for n-C7 H16 /O2 /Ar (1) The additional relative errors on the modeling,
generated by the correlation, remain negligi-
4.1. General ble compared to the detailed mechanism inac-
curacies: the discrepancies (simulation τsim |
This section and the following one describe the experience τexp are larger than 25%, that, is more
general evolution tendencies of the modified mech- than one order of magnitude higher than the stud-
anism predictions. The study of the ignition delays ied deviations (τcor | τsim ).
396 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

Fig. 5. Evolution of the correlated ignition delays τ cor , with the equivalence ratio Φ and the temperature T for the mixture
n-C7 H16 /O2 at P ref = 100 kPa: (a) 3D representation of τcor = f (Φ, T ) over the investigated range (log–normal–log scale);
(b) cross sections τcor = fT (Φ) of the cartography, realized at 1500 K, 1600 K, 1700 K, and 1800 K (log–log scale); (c) cross
sections of the cartography, carried out according to the valley: illustration of the evolution with T of the minimum delay τ cor, min
(normal–log scale) and of the associated optimal equivalence ratio Φopt,τ (normal–log scale). Use of Eqs. (A.21) and (A.22),
reproducing the predictions of the modified mechanism (Imbert et al. [1]) with an average deviation about 0.6%.

(2) The absolute errors are too small to deteriorate 4.2. n-C7 H16 /O2 /Ar mixtures diluted and
the graphics intended to reproduce the evolution stoichiometric
tendencies of τ sim : τ cor constitutes a very good For stoichiometric mixtures, either pure or diluted
approximation of τ sim , due to the important vari- in argon, the evolution of τ cor with T , P , and X Ar
ations of τ sim inside the studied range of condi- does not require any particular comments in addition
tions. to those formulated after the two-dimensional analy-
sis. The reason is the important decoupling between
Taking advantage of these features, the evolution these three kinetic factors: as exposed in Tables 4
of the ignition delays τ cor and of the activation ener- and 5, the absolute values of the crossed coefficients
gies θ a, cor will be often represented as tridimensional are quite small.
isometric plots here. Indeed, illustration of the evolu-
tion tendencies of these parameters enables the dis- 4.3. n-C7 H16 /O2 mixtures pure and with variable
cussion of the validity domain of the mechanism: for equivalence ratio
example, Davidson and Hanson [13] used this method The conclusion is not the same for pure mixtures
to compare the delays issued from the semi-empirical with variable stoichiometry, as can be shown by the
correlations with those predicted by the mechanisms evolution of τcor = f (Φ, T ) (Fig. 5). In contrast to
they studied in the case of the combustion of n- the previous case, the thermodynamic parameters (T
heptane and iso-octane. or P ) definitely overlap with the equivalence ratio
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 397

ones, especially for lean mixtures. To quantify this θ a, cor and E a, cor were evaluated after the inter-
characteristic, Fig. 5b illustrates cross sections τcor = mediate calculation of three delays τ cor , established
fT (Φ) carried out for various constant temperatures. with Eq. (31):
This graphic shows:
• for the desired operating conditions (P , Φ, and
• a variation with Φ of the gap between two sec- X Ar );
tions established at fixed T , which manifests the • at three arbitrary chosen temperatures: 1500,
coupling between the parameters; 1700, and 1900 K.
• a minimum delay τ sim, min and an associated op-
timal equivalence ratio Φopt,τ ≈ 0.3, both pa- Ea,cor = R · θa,cor represents the slope of the line
rameters varying with the thermodynamic state ln(τcor ) = f (T −1 ), which best correlates these val-
(Fig. 5c). ues in an Arrhenius-type diagram.
As E a, cor originates from a derivative of τ cor ver-
The optimal kinetic equivalence ratio Φopt,τ in- sus T , particular care has been taken to also check the
creases with T : at higher temperatures, the competi- continuity of the derivative of τ cor versus Φ. This has
tion between RH and O2 to react with ◦ H has less been the case especially, for the connection between
impact, with regards to the greater production of this the mixtures (Section 3.5):
atom. However Φopt,τ remains rather far from the op-
timal thermodynamic one Φopt,DCJ ≈ 2.2 (Imbert et • rich: quasi-linear dependence of Lτsim =
al. [4]), associated with the maximum of: fT ,P (LΦ);
• lean: parabolic evolution of Lτsim = fT ,P (LΦ).
• the detonation velocity DCJ ≈ 2400 m s−1 ≈
8640 km h−1 ≈ 5370 mph, corresponding to The particular features, hereafter exposed, will not
Mach 9.4 (whereas for the stoichiometric mixture be interpreted: this exercise leaves the framework of
DCJ ≈ 2100 m s−1 ≈ 7560 km h−1 ≈ 4700 mph the semi-empirical correlations reformulation and ap-
⇔ Mach 7.2); pears difficult.
• the von Neumann temperature T vN ≈ 1700 K, As for the analysis of the previous coefficients, a
which is the average of the extremely heteroge- good interpretation would require:
neous conditions behind the detonation front;
• the detonation intensity PPvN − 1 ≈ 100. • a thorough theoretical study of the mechanism;
1 • an experimental investigation to establish a ki-
This discrepancy between kinetic and thermodynamic netic database, statically broad enough to smooth
the evolution trends of the measured delays τ exp ,
constraints qualitatively explains why the highest sen-
and then to calculate trustable local derivatives
sitivity to detonation corresponds to slightly rich mix-
E a, exp .
tures (Φ ≈ 1.1, according to Imbert et al. [4]).
Incidentally, the normal scale, reported along the
5.2. n-C7 H16 /O2 mixtures pure and stoichiometric
temperature axis, does not constitute the best choice,
as clearly seen by fixing Φ: τ cor , if plotted versus T
Fig. 6 illustrates the evolution of θa,cor = f (P ),
using a logarithmic scale, is submitted to relatively
for the reference mixture. For a given composition,
important inflections. A scale according to T −1 (as
θ a, cor grows with P and seems to tend toward a max-
in a standard Arrhenius-type diagram) would be more
imum θa,cor,max ≈ 19.6 × 103 K (E a, cor, max ≈ 163
suitable for the temperature axis.
kJ mol−1 ≈ 39.0 kcal mol−1 ), which is approached in
the vicinity of P opt ≈ 1 MPa = 10 bar ≈ 145 psi.
5. Evolution of the activation parameters θa,sim
5.3. n-C7 H16 /O2 /A mixtures diluted and
and Ea,sim , simulated for n-C7 H16 /O2 /Ar stoichiometric
5.1. General Fig. 7 illustrates the evolution of θa,cor =
f (P , XAr ) of the stoichiometric mixture n-C7 H16 /
This part describes the evolution of the appar- O2 /Ar. θ a, cor seems:
ent activation factors (temperature θ a, sim and energy
E a, sim ) with: • to tend toward the maximum θa,cor,max ≈ 19.6 ×
103 K located close to P opt = 1 MPa, for a given
• the mixture nature (Φ and X Ar ); composition (here fixed X Ar );
• the initial pressure P of the shocked medium. • to differently evolve with X Ar according to P :
398 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

Fig. 6. Evolution of the correlated activation temperature


θ a, cor with the pressure P (log–normal scale) for the mix-
ture n-C7 H16 /O2 , Φref = 1. Use of Eq. (A.19), reproducing
the predictions of the modified mechanism (Imbert et al. [1])
with an average deviation about 0.3%.

Fig. 8. Evolution of the correlated activation temperature


θ a, cor with the equivalence ratio Φ and the pressure P for
the mixture n-C7 H16 /O2 : (a) 3D representation of θa,cor =
f (Φ, P ) over the investigated range (log–log–normal scale);
(b) cross sections θa, cor = fP (Φ) of the cartography real-
ized at 1.0 MPa, 316 kPa, 100 kPa, 31.6 kPa, and 10.0 kPa
(log–normal scale). Use of Eqs. (A.21) and (A.22), repro-
ducing the predictions of the modified mechanism (Imbert
Fig. 7. Evolution of the correlated activation temperature et al. [1]) with an average deviation about 0.6%.
θ a, cor with the inert fraction X Ar and the pressure P
(normal–log–normal scale) for the mixture n-C7 H16 /O2 /Ar,
Φref = 1. Use of Eq. (A.20), reproducing the predictions of
the modified mechanism (Imbert et al. [1]) with an average • to increase with P and then to decrease for a
deviation about 0.3%. given composition (here fixed Φ): for rich mix-
tures (Φ  1.25), the observed maximum is at-
tained for P  103 kPa;
• to reach its maximum at Φ ≈ 1, by varying Φ
– θ a, cor decreases with X Ar for low pressures and fixing P ∈ [36 kPa, 900 kPa] = [360 mbar,
(P ≈ 10 kPa); 9.0 bar] ≈ [5.2 psi, 130 psi] (Fig. 8b): it appears
– θ a, cor seems not very sensitive to dilution for as a crest line (well approximated in Fig. 6, re-
high pressures (P ≈ 103 kPa). lated to the stoichiometric mixture).

5.4. n-C7 H16 /O2 mixtures pure and with variable This inclination change (discontinuity of the second
equivalence ratio derivative of τ cor , first established from T to form
θ a, cor , then from Φ) would remain undetectable with-
Fig. 8 illustrates the evolution of θa,cor = f (P , Φ) out the use of a systematic calculation technique, as
of the mixture n-C7 H16 /O2 . In Fig. 8a, θ a, cor seems: outlined here. This particularity cannot be entirely at-
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 399

tributed to Eqs. (A.21) and (A.22), since these corre- In Appendix A, the mathematical way to build and
lations reproduce the simulated delays τ sim with an optimize this new formalism, as well as characterize
average deviation close to 1.5%. This characteristic its performances, was explained.
seems to result from an artifact of the detailed mech-
anism, not interpreted here for the previously men- (1) The correlation of τ sim by τ cor , thanks to a gen-
tioned reasons. eralized development in Taylor series [36], was
proved to be meaningful using simultaneously
the four factors T , P , X Ar , and Φ:
6. Conclusion and perspectives: Potentialities of   2 2 2
the advanced correlations τ
lg = LΦ k · LXj · LPi
τref
k=0 j =0 i=0
The main motivation concerning this article was
the rapid, easy, and precise evaluation of the ignition × (ai,j,k + bi,j,k · IT + ci,j,k · LT)
delays, from global parameters describing the ther- (Eq. (31)).
modynamic state, and the composition of the reactive
(2) A kinetic database, composed of 5N delays for
mixture. The method, a particular adaptation of the re-
N ∈ [1, 4] parameters and quickly prepared was
promodeling philosophy, was developed considering
proved to be large enough if efficiently designed
the particular case of high-temperature combustion
using strategic conditions.
(occurring behind shock waves or detonation fronts)
(3) The coefficients τref , . . . , ci,j,k (here expressed
of n-heptane, supposed to be a good model fuel for
for N = 4), solutions of a quasi-linear over-
PDE.
sized equation system, can be rapidly and ex-
These specificities explain why the work presented
actly determined, applying the Moore and Pen-
here was inspired by the traditional semi-empirical
rose pseudo-inverse [35], a no-iterative analytical
correlations, which have recognized abilities to repro-
process.
duce ignition delays.
(4) A way to speed up the evaluation of the poly-
In Section 2, the fits τ cor were interpreted:
nomials simply by organizing the calculation to
reduce its complexity was explained.
(1) by reformulating them, starting from the param-
(5) The orders of magnitude of the durations re-
eters T , P , Φ, and X Ar ;
quired to evaluate the traditional correlation, the
(2) by explaining the four dependence types between
new formalism, and to use S ENKIN with the de-
the delays τ and these kinetic factors, using well-
tailed mechanism, were quantified and then com-
established theoretical models: Arrhenius [14],
pared.
Lindemann et al. [23], etc.;
(6) The typical deviations between the simulated de-
(3) by physically analyzing the coefficients θ a, cor ,
lays τ sim and their fits τ cor were evaluated.
α cor , p, and x deduced from these relations.
Last, using Eq. (31), some general features of the
Then in Section 3, the transposition of this classi-
mechanism were exhibited in Sections 4 and 5:
cal formalism to a detailed mechanism was discussed.
The analysis of the impact of T , P , X Ar , and Φ on the
(1) The delay τcor = f (T , P , Φ, XAr ) evolves in a
delays τ sim , simulated by the version of the Curran et
complex way with Φ, due to the strong coupling
al. model [6,7], slightly modified by Imbert et al. [1],
of this parameter with the thermodynamic state
allowed us:
(T or P ).
(2) The activation temperature θa,cor = f (P , Φ,
(1) to exhibit similar evolutions for the mechanism
XAr ) exhibits an unexpected maximum around
and the semi-empirical correlations: only the de-
stoichiometry, marked by an inclination change.
pendence {τsim | Φ} differs, especially for fairly
This observation remains difficult to experimen-
lean mixtures;
tally confirm, other than indirectly by deriving
(2) to find good agreement between the coefficients
the order of magnitude of the measured delays
θ a , α, p, and x, deduced both from the semi-
Ln(τexp ) versus T −1 .
empirical correlations and from the mechanism;
(3) to define five reduced variables (cf. Section 2.6),
The performances of these new correlations were
intended to be kinetically significant:
checked with ignition delays τsim = f (T , P , Φ, XAr ),
• IT = 1− TTref , LT = lg( TT ), and LP = lg( PP ) simulated by a mechanism and not empirically deter-
ref ref
for the thermodynamic state; mined, due to the lack of measurements to establish
1−XAr
• LX = lg( 1−X ) and LΦ = lg( ΦΦ ) for the an experimental kinetic database, broad and diversi-
Ar,ref ref
mixture nature. fied enough.
400 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

The polynomial interpolations τ cor seem satisfac- used during the development phase, or the validation
tory, since they reproduce the delays τ sim with a good stage of a mechanism:
precision in comparison with the reliability to predict
τ sim using the original detailed mechanism: in the (1) to highlight and to compare the general evolution
most unfavorable case, the 3.5% deviation of {corre- tendencies of the delays:
lation τ cor | reference τ sim } appears lower, by almost • τ exp , empirically measured (prerogative tra-
one order of magnitude, than the uncertainties ris- ditionally dedicated to the semi-empirical
ing from simulations (discrepancies {modeling τ sim Arrhenius-type correlations);
| experiment τ exp } estimated at more than 25%, ac- • τ sim , simulated by a kinetic model (as here,
cording to Imbert et al. [1]). with n-heptane oxidation);
In addition, this method is really flexible to use, (2) to notice, with an already built mechanism, cer-
considering: tain troubles difficult to be detected without a tool
dedicated to the systematic estimation of τ sim :
• the number of kinetic factors N ∈ [1, 4] (here se- crest-line of θa,cor = f (Φ, P ), for example;
lected among T , P , Φ, and X Ar ), likely to be (3) to facilitate the targeting of operating conditions
simultaneously investigated; in an experimental device (shock tube or rapid
• the broad range of investigated conditions (no ex- compression machine) by predicting strategic
haustive values, here chosen for a later kinetic ones to validate a model thanks to τ exp .
analysis of n-heptane detonation), resulting from
the combination of the domains: In the case of the modeling of the high-temperature
– T ∈ [1500 K, 1900 K]; combustion, the suggested method can also constitute
an alternative to the classical mechanism reduction,
– P ∈ [10 kPa, 1 MPa];
since it conciliates:
– XAr ∈ [0, 0.7];
– Φ ∈ [0.25, 4.0].
(1) the powerful data processing of both empirical
correlations and global reaction models, tech-
Moreover, the computational speed up is:
niques largely employed today in computational
simulations, due to the intensive calculations of
• important: computation about 3 × 106 faster than
reaction rate progress, related to mesh grids with
the step-by-step resolution of the numerous dif-
a significant number of nodes;
ferential equations, using S ENKIN;
(2) the precision and flexibility of the detailed mech-
• comparable with the traditional semi-empirical
anisms, provided that the “freezing” of the influ-
correlation one: calculation time only 13 times
ence of the neighborhood on the chemistry dur-
longer. ing the autoignition does not constitute a severe
oversimplification;
This noticeable gain does not really depend on the (3) the ease and rapidity of use, which make it quasi-
computer power, but on the complexity of both the directly available to researchers from all special-
correlation and the mechanism. The performances ities, with little effort and without the need of
have been evaluated here for a polynomial with two specific tools.
parameters (i.e., 32 = 9 coefficients), reproducing a
detailed model composed of 565 species and 2538 re- In particular, this new polynomial expression can
actions. be perfectly integrated in the modeling of the com-
Last, from a technical point of view, this construc- bustion behind a shock front, since it can be em-
tion method respects two other quality modeling stan- ployed:
dards:
(1) directly with the traditional ZND model for a
• retaining the physical meaning: fidelity to the global “0D” interpretation of the kinetic proper-
empirical Arrhenius rate law [14] and to its adap- ties of a CJ-type self-sustained detonation;
tation by Lifshitz et al. [16], for example; (2) as calculation intermediaries to evaluate the re-
• limiting the number of adjustment parameters action rate progress, in multidimensional com-
(3N  81 coefficients, here for N ∈ [1, 4]), plex processes (detonation diffraction, transition
and offering the possibility to interpret these from deflagration to detonation, . . . ): the de-
terms. lays remain so short (few microseconds) that
mass and energy transfers (otherwise than in-
Because of its interpolation quality in a wide range duced by shock wave complexes) can be ne-
of conditions, this new type of correlation can first be glected.
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 401

These remarks emphasize the limitations of this Appendix A. Development of the polynomial
process, which remain comparable with those of the correlations and calculation of the related
semi-empirical correlations and of the behavior mod- performances and uncertainties
els, generally speaking:
A.1. General method to extend the Arrhenius-type
(1) Lack of modularity: as with “purely mathemati- formalism
cal” models, this black box only reproduces one
fixed image of the original kinetic model, which A.1.1. Correlation with fixed coefficients
must have been previously validated. However, The correlation with fixed coefficients constitutes
the wasted time resulting from a mechanism re- the simplest Arrhenius-type expression to be built,
vision remains modest taking into consideration starting from the parameters IT, LT, LP, LX, and LΦ:
the profit likely to rise from its later intensive use  
in CFD; τcor
Lτ = lg
(2) Loss of precision while extrapolating far from τref,0
the domain related to the database establish-
= −θ · IT + t · LT + p · LP + p · LP2
ment;
(3) “Freezing” of chemistry: the mass transfer seems + x · LX + x  · LX2 + φ · LΦ + φ  · LΦ 2 .
to prohibit the use of this means to simulate tra- (A.1)
ditional engines or turbines. Since temperatures As the time constant is expressed inside lg(τref,0 ), the
attained in deflagrations are lower, the involved logarithm of the dimensionless duration Lτ quite nat-
reaction times, of about several milliseconds, al- urally provides the reaction delay, according to:
low the displacements and the mixing of active
species over distances not negligible compared τ = τref,0 · 10Lτ ⇔ (A.2a)
to typical sizes of the enclosure, and so of the  
mesh grid. τcor
Lτ = lg = f (T , P , Φ, XAr ). (A.2b)
τref,0
More elaborate models (based on this polynomial
Expression (A.1) constitutes only a slight adaptation
formalism) could reproduce the kinetic impact of a
of the linearized version of the semi-empirical cor-
thermodynamic change due to external reasons during
relation (Eq. (12)). The addition of second-order cor-
the induction process. So the rapid variation of tem-
rective terms and the reformulation of the dependence
perature and pressure generated by the shock waves
{τcor | Φ} obviously improve the capacities to repro-
propagation could be taken into account.
duce the reference delays τ sim . However, the posi-
A notable improvement to definitely break the
tive impact only remains localized in the vicinity of
“freezing” of the autoignition chemistry would thus the development center (here related to the conditions
consist of taking into account the influence of the chosen as references).
neighborhood, concerning concentrations of active
species.
A.1.2. Local development around the center of the
This limitation currently restricts the use of this
domain
correlation to hot combustions, very fast compared
Correlation (A.1) suffers from the non-adaptability
to the mass transfers (so with a Damköhler number
of its coefficients. In fact, to preserve a good repro-
Da 1), like those related to PDE. By aiming at this
duction quality, it would be necessary:
objective, it may be possible in the future to take into
account the external modifications of key reactional
• either to strongly restrict the applicability range,
concentrations (related to the pool of atoms and rad-
as do the majority of the users of semi-empirical
icals active during the propagation stage). The ideal
correlations;
would then be to manage to build independent re-
• or to vary the coefficients, versus parameters they
duced variables, able to be easily incorporated into
do not directly correlate with; for example, even
polynomial expressions similar to the one presented
if the simulations show a slow dependence, the
here.
pressure coefficient p should slightly evolve with
T , X Ar , and Φ.

Acknowledgments By aiming at this last point, it is possible to develop


Eq. (A.1), according to the four considered factors
The authors thank Dr. Henry Curran, from NUI (T , P , X Ar , and Φ). The use of the Taylor series
Galway, for supplying his detailed kinetic model. [36], limited to the vicinity of the reference condition
402 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

τref = f (XAr,ref , Φref , Pref , Tref ), provides: • limiting the calls of time-consuming functions:
  namely inverses, logarithms, and powers related
τcor to real exponents;
Lτ = lg
τref • using an appropriate factorization, i.e., replacing
= [1 + a0 · LΦ + a0 · LΦ 2 ] the sum of terms containing powers bounded to
entire exponents, by a series of simple additions
× [1 + b0 · LX + b0 · LX2 ] and multiplications.
× [1 + p0 · LP + p0 · LP2 ]
Consequently, the first step consists of determining
× [1 + t0 · LT − θ0 · IT] − 1. (A.3) the reduced variables, which evaluations are impossi-
ble to avoid (among IT, LT, LP, LX, and LΦ), once for
A.1.3. Generalization of the development to the each condition. In principle, it is then enough to es-
whole domain tablish the correlation, while taking as a starting point
The development of the Taylor series, written ac- the following example. By supposing that a second-
cording to the factorized form (Eq. (A.3)) exhibits order polynomial correctly reproduces each evolution
terms additional to those shown by relation (A.1): the {τ sim | kinetic factor}, it is a matter of evaluating
new crossed coefficients take into account a coupling the following line vectors V 1,n . With their N com-
between each parameter. The overlap of the param- ponents, they can exhibit the simultaneous influence
eters {RT or LT}, LP, LX, and LΦ complicates the of:
polynomial, but partially meets the formulated need
for flexibility. It is possible to improve the adaptabil- • T:
ity by suppressing the constraints implicitly imposed
by factorization with: VT1,3 = ( 1 IT LT ) ; (A.4a)
  • T and P :
τcor
Lτ = lg  
τref,0 VTP1,9 = VTP(0) V (1)
TP1,3
V
TP1,3
(2) , (A.4b)
1,3

2
2
2
with:
= LΦ k · LXj · LPi
k=0 j =0 i=0 V = VTP1,3 , (A.4c)
(0)
TP1,3
× (ai,j,k + bi,j,k · IT + ci,j,k · LT)
V (1) = LP · V (0) , (A.4d)
TP1,3 TP1,3
(Eq. (31)).
V (2) = LP · V (1) ; (A.4e)
Here the implicit time constant τ ref, 0 (deduced from TP1,3 TP1,3
the coefficients resolution) borders τ ref , the delay di- • T , P , and X Ar :
rectly predicted for the reference condition. The re-  
production of τ sim remains good not only in the vicin- VTPX = V
1,27 TPX1,9
(0) V
TPX1,9
(1) V
TPX1,9
(2) ,
ity of the development center (as for any Taylor se- (A.4f)
ries), but also in the whole range, due to the explicit
with:
generalization of the coupling between parameters.
If the number of coefficients 3N (for N parame- V (0) = VTP1,9 , (A.4g)
ters) appears too important, for an easy and fast han- TPX1,9
dling, it seems necessary to subdivide the correlation V (1) = LX · V (0) , (A.4h)
TPX1,9 TPX1,9
validity range. This simplification enables the preser-
vation of good agreement with the detailed mecha- V (2) = LX · V (1) ; (A.4i)
TPX1,9 TPX1,9
nism, while limiting the inflation of adjustment pa-
• T , P , X Ar , and Φ:
rameters.
VTPXΦ1,81
A.2. Optimization of the polynomial evaluation  
= VTPXΦ (0) V
TPXΦ1,27
(1) V
TPXΦ1,27
(2) ,
1,27
A.2.1. Theoretical principle (A.4j)
The evaluation of the correlated delay τ cor can be with:
accelerated on computer by decreasing the number of
complex operations necessary to its calculation. This V (0) = VTPX1,27 , (A.4k)
TPXΦ1,27
traditional algorithm in data processing here consists
of: V (1) = LΦ · V (0) , (A.4l)
TPXΦ1,27 TPΦ1,27
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 403

V (2) = LΦ · V (1) . (A.4m) • [Ar]0.09 not taken into account;


TPXΦ1,27 TPXΦ1,27
• [C7 H16 ] and [O2 ] previously established starting
These stages finally enable the building of V TPXΦ 1,81 , from X C7 H16 , X O2 , T , and P , with Eq. (3).
representative of the mixture conditions:
So the calculation time related to the new formalism
VTPXΦ1,81 = (1 IT LT LP LP · IT appears:

LP · LT LP2 LP2 · IT ... • only 13 times longer than the one bounded to the
LΦ 2 · LX2 · LP2 · LT). (A.5) classical correlation;
• about 3×106 times shorter than the one linked to
The scalar product of this line vector with the col- the simulation using C HEMKIN.
umn vector C 81,1 (composed of the polynomial coef-
ficients) provides Lτ , the logarithm of the correlated Obviously, this important improvement depends on
delay τ cor , scaled by τ ref, 0 : the complexity of:
 
τcor
Lτ = lg = VTPXΦ1,81 · C81,1 . (A.6) • the polynomial, which decreases with the number
τref,0 of coefficients (3N rising itself with the number
N ∈ [1, 4] of kinetic factors);
A.2.2. Performance analysis
• the original mechanism, which increases with the
To quantify the evaluation duration of the corre-
number of species and reactions.
lated delay τ cor , these vectorial operations have been
carried out:
A.3. Analytical resolution of the polynomial
coefficients
• with a Pentium III running at 500 MHz;
• on M ATLAB 6.1 (interpreted language, especially
dedicated to matrix algebra); A.3.1. Theoretical principle
• by considering a correlation with 9 coefficients This passage shows how to optimize the 3N coef-
(restricted to 2 parameters: T and P ); ficients (noted from a 0,0,0 to c2,2,2 ), that is, to min-
• while taking as previously established parame- imize the difference between the delays τ sim (sim-
ters: ulated by the mechanism and used as references)
– T = 1800 K ≈ 1525 ◦ C ≈ 2780 ◦ F; and τ cor (evaluated with this new correlation), for
– P = 200 kPa = 2.0 bar ≈ 29 psi; the whole conditions described by an existing kinetic
– the polynomial coefficients, and among them: database.
a0,0,0 = lg(τref,0 ). Using the formalism of Eq. (31), this condition
corresponds to solving the following oversized equa-
Under these conditions, 107 loop repetitions have en- tion system (i.e., with more constraints than unknown
abled the estimation of the total duration at 308 µs, as factors):
well as the percentages of the computing times dedi-
⎧ 2 2 j
cated to each stage: ⎪ 2 k i

⎪ k=0 j =0 i=0 LΦ1 · LX1 · LP1



⎪ × (ai,j,k + bi,j,k · RT1 + ci,j,k · LT1 )
• evaluation of the variables IT, LT, and LP (Eqs. ⎪


⎪ ≈ lg(τsim,1 ),
(26)–(28)): 174 µs ⇔ 56.6%; ⎪


⎪ 2 2 2
• determination of the line vector V TP1,9 , by com- ⎪
⎪ k=0 j =0 i=0 LΦ2k · LX2 · LPi2
j


bination (Eqs. (A.4a)–(A.4e)): 98 µs ⇔ 32.0%; ⎪

• calculation of the scalar product (Eq. (A.6)): × (ai,j,k + bi,j,k · RT2 + ci,j,k · LT2 )
26 µs ⇔ 8.4%; ⎪ ≈ lg(τsim,2 ),




⎪ ..
• passage to the exponent according to τcor = 10Lτ ⎪
⎪ .


(Eq. (A.2)): 9.0 µs ⇔ 3.0%. ⎪
⎪  2 2

⎪ 2 k j i

⎪ k=0 j =0 i=0 LΦM · LXM · LPM


The estimation of τ sim using S ENKIN, by an iterative ⎪
⎪ × (ai,j,k + bi,j,k · RTM + ci,j,k · LTM )

resolution of the whole differential equations associ- ≈ lg(τsim,M ). (A.7)
ated to the detailed mechanism (2538 reactions and
565 species), was approximately 15 min. By adopting as a criterion the least-squares tech-
The time needed to evaluate Imbert et al. semi- niques, and weighting each of the M terms by Wl ,
empirical correlation (Table 2) was 24 µs, under sim- the objective function to be optimized can be written
ilar calculation conditions: as
404 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

f (a0,0,0 , . . . , c2,2,2 ) Matrix


Weighted mixture description: thermodynamic state
 2 2 2

M j
and composition
= LΦlk · LXl · LPil ⎛√ ⎞
W1 · V (1)
l=1 k=0 j =0 i=0 TPXΦ1,81
⎜ √ ⎟
⎜ W2 · V (2) ⎟
⎜ TPXΦ1,81 ⎟
× (ai,j,k + bi,j,k · RTl + ci,j,k · LTl ) MM,81 = ⎜ ⎟. (A.11)
⎜ .. ⎟
⎜ . ⎟
2 ⎝√ ⎠
WM · V (M)
− lg(τsim,l ) · Wl . (A.8) TPXΦ1,81
Column vector
Delays τ sim simulated by the model and weighted
Several numerical optimization methods would
⎛ √ ⎞
√W1 · lg(τsim,1 )
enable the solving of system (A.8):
⎜ W2 · lg(τsim,2 ) ⎟
⎜ ⎟
• analytical ways, generally based on the gradients: TM,81 = ⎜ .. ⎟. (A.12)
⎝ . ⎠
improvement of Newton and Raphson’s original √
principle [37,38]; WM · lg(τsim,M )
• heuristic processes (also known as stochastic or Column vector
geometrical): exploration of the space by suc- Polynomial coefficients to be solved (notation
cessive tests and temporary assumptions to re- Eq. (31))
search the most favorable directions: the Nelder ⎛ ⎞
a0,0,0
and Mead simplex [39], for example; ⎜ b0,0,0 ⎟
• metaheuristic means, improvements of the previ- ⎜ ⎟
⎜ ⎟
ous methods, employed to solve the difficult arti- C81,1 = ⎜ c0,0,0 ⎟ . (A.13)
⎜ . ⎟
ficial intelligence problems: the Schwefel evolu- ⎝ .. ⎠
tion strategy [40], for example. c2,2,2
This linear optimization problem can be directly
The choice of an optimization process is not without solved using the analytical method developed by
consequence, since it can: Moore in 1920 and published by Penrose [35].
This matrix algorithm, based on the evaluation of
• converge only at the end of an important series of
iterations, and so notably lengthen the calculation • Moore’s pseudo-inverse matrix:
time;
• block on a simple local optimum, and thus not †† M t −1 · t M
81,N = ( MN,81 · MN,81 ) N,81 ;
provide the expected global minimum;
(A.14)
• sometimes require a delicate adjustment of the
initial conditions, to avoid the previous nui- • the vector solution:
sances.
C81,1 = †† M81,N · TN,1 ; (A.15)
In this case, as equation system (A.7) is intentionally
gives in less than one second the exact solution when
quasi-linear, the research of least-squares becomes it is programmed with M ATLAB (the Moore pseudo-
very simple. It is enough: inverse can be evaluated by the function pinv).
Note: As Tables 4–7 show, the polynomial con-
• to minimize stant (first component of the column vector C81,1 ) is
not written as it was originally, but retranscribed in
f (a0,0,0 , . . . , c2,2,2 ) the more explicit form of a pseudo-reference delay

= MN,81 · C81,1 − TN,1 2 ; (A.9) τref,n = 10a0,0,0 . (A.16)


τref,n appears very close to the delay τ ref simulated
• that is, to obtain
using C HEMKIN, for the reference conditions.

MN,81 · C81,1 ≈ TN,1 , (A.10) A.3.2. Choice of the kinetic database


Detailed mechanism characteristics were recorded
with the following elements: in the form of a base of M data, including:
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 405

• the thermodynamic state of the reactive medium • Φ = 1.0.


(RT, LT, and LP established from T , P ) to evalu-
ate MM,81 (matrix (A.11)); A.4. Evaluation of the deviation
• the nature of the initial mixture (LX and LΦ eval- {correlation τcor | simulation τsim }
uated from X Ar and Φ) to determine MM,81 ;
• the simulated times (Lτ established from τ sim ) Here, the average dispersion {correlation τcor |
resulting from the choice of these conditions reference τ sim } will not be expressed using the Pear-
and employed to calculate TM,81 (column vec- son correlation coefficient χ 2 [41]. Indeed, as the
tor (A.12)). analysis relates to a comparison between orders of
magnitude in time, for a fast interpretation, it seems
To shorten the resolution stage of the polynomial more explicit to use a definition inspired by the stan-
coefficients, it appears possible to restrict the database dard geometrical deviation:
size to M = 5N key values (for N parameters). These

selected conditions should correspond to grid nodes:  N   τcor (i) 2 
i=1 ln τsim (i)
Gτcor = exp
• regularly spaced, within sight of the typical ki- N
netic evolution scales, for each parameter (see
 1. (A.17)
the definition of the reduced variables IT, LP, LX,
and LΦ) (Eqs. (26)–(30)); The reproduction quality by the polynomial correla-
• astutely placed to minimize the distance between tions remains good: Gτcor  1.05 is verified for the
two consecutive nodes by taking into considera- four explicit correlations, when Gτcor = 1 corre-
tion a suitable norm for the studied space, which sponds to the equality between τ cor and τ sim . Con-
dimension is N . sequently, the discrepancy can be expressed by a rel-
ative deviation, more natural:
Then, the weights Wl of each condition depend on:
τcor = 100(Gτcor − 1)  5%. (A.18)
• the prediction reliability, which is here nearly the
same for all the delays τ sim , due to:
– the good coverage between the mechanism va- A.5. Characteristics of the polynomial correlations
lidity range and the domain conditions used; in the case of n-heptane autoignition
– the retention of the tolerance to evaluate τ sim
(defined by the time interval DELT for solu- A.5.1. General
tion printouts, in the C HEMKIN input file); This part clarifies the general correlation (Eq. (31))
– the automatic procedure to determine τ sim , in the particular cases of four investigation ranges de-
starting from the time profile of [◦ OH]; fined to analyse the kinetic properties related to the
• the strategic position of the chosen condition, detonation of n-heptane/oxygen mixtures:
within the mesh grid: the weighting (left to the
arbitrary choice of the user) is worth here: • not diluted and stoichiometric: XAr = 0 and
– 3, for the extreme or central positions, accord- Φ = 1;
ing to the considered kinetic scale; • not diluted and rich: XAr = 0 and Φ ∈ [1.0, 2.5];
– 2, for intermediate, and regularly spaced, po- • not diluted and lean: XAr = 0 and Φ ∈ [0.25,
sitions; 1.0];
– 1, for randomly generated conditions. • diluted in argon and stoichiometric: X Ar ∈
[0, 0.7] and Φ = 1.
Once these precautions are taken, the resolution
time of the polynomial corresponds to the duration This division of the investigation range has enabled
needed to prepare the database. Thus to analyse n- the limitation of the total number of coefficients, with-
C7 H16 /O2 /Ar mixtures, 53 × 15 = 1875 min (31.5 h, out deteriorating the reproduction quality, on the large
or 1.3 days) is needed, using a Pentium III, 500 MHz). domain related to the thermodynamic state:
These calculations enable the reproduction of the sim-
ulated delays τ sim , with an average deviation of 0.3%, • T ∈ [1500 K, 1900 K];
over the broad domain simultaneously covering: • P ∈ [10 kPa, 1 MPa].

• T ∈ [1500 K, 1900 K]; A.5.2. n-C7 H16 /O2 mixtures pure and
• P ∈ [10 kPa, 1 MPa]; stoichiometric
• XAr ∈ [0, 0.7]; Table 4 lists the coefficients of the polynomial
406 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408
 
τcor + LP · (u2 + v2 · IT + w2 · LT)
lg = (a1 · IT + b1 · LT)
τref,1 
+ LP2 · (x2 + y2 · IT + z2 · LT) .
+ LP · (c1 + d1 · IT + e1 · LT)
(A.20)
+ LP2 · (f1 + g1 · IT + h1 · LT).
Correlation (A.20) provides the delays τ cor associ-
(A.19) ated with the mixture (n-C7 H16 /O2 /Ar, Φ = 1.0, X Ar
Correlation (A.19) provides the delays τ cor associ- ∈ [X Ar, ref = 0, 0.7]). The relative deviation {τcor |
ated with the mixture (n-C7 H16 /O2 , Φ = 1.0). The τsim } is lower than 1.5% (0.3% on average) for:
average deviation {τcor | τsim } compared to the model
predictions is lower than 1.5% (0.3% on average) for: • 73 time values τ sim a priori used to establish the
database;
• 15 time values τ sim a posteriori evaluated for
• 29 time values τ sim a priori used to establish the
random conditions.
database;
• 10 time values τ sim a posteriori evaluated for
The coefficients analysis shows that:
random conditions.
(1) the terms τ ref, 2 , a2 − h2 remain almost the same
Due to the slow and regular evolution of the delays as their counterparts (τ ref, 1 , a1 − h1 ) related to
τ sim with T and P , it appears reasonable to assume the reference composition, a consequence of the
that the relative deviation between the simulations lost influence of the terms i2 − z2 (related to the
τ sim and their correlations τ cor remains lower than powers in LX) when the mixture is no longer di-
1.5% for the whole range of conditions. luted with argon (XAr = 0 and LX = 0);
The coefficients analysis shows that: (2) the higher order terms j2 − q2 and s2 − z2 (re-
lated to LX and LX2 ) often remain smaller than
(1) τ ref,1 remains very close to τref = 5.95 µs de- their counterparts a2 − h2 of the reference mix-
lay predicted by the mechanism for the refer- ture, a result of the lost influence of the corrective
ence condition (Tref = 1700 K, Pref = 100 kPa, terms, related to the high-order powers of LX.
Φref = 0 and XAr,ref = 0);
(2) the high absolute value of the coefficients a1 − b1 These properties are typical of the developments in a
(related to LT) compared to c1 − h1 (bounded Taylor series, limited around a central point, to which
to LP) confirms the greater influence of the rela- the function converges. In this case, these character-
tive fluctuations of T (than those of pressure P ) istics are remarkable due to the selected formalism:
around the central position; thus, the importance of optimizing the form of the
dimensionless kinetic variables, in order to obtain a
(3) the coefficients d1 − h1 , associated with the high-
rapid convergence.
order terms (products of IT, LT and LP), remain
lower than a1 − b1 , as a consequence of the
A.5.4. n-C7 H16 /O2 mixtures pure and with variable
weak influence on Lτ of the corrective terms
equivalence ratio
intended to reproduce the correlation curvatures
Tables 6 and 7 list the coefficients of the polyno-
(Sections 3.1 and 3.2).
mials
 
τcor
A.5.3. n-C7 H16 /O2 /Ar mixtures diluted and lg = (a3 · IT + b3 · LT)
τref,3
stoichiometric
+ LP · (c3 + d3 · IT + e3 · LT)
Table 5 lists the coefficients of the polynomial
+ LP2 · (f3 + g3 · IT + h3 · LT)
  
τcor
lg = (a2 · IT + b2 · LT) + LΦ · (i3 + j3 · IT + k3 · LT)
τref,2
+ LP · (l3 + m3 · IT + n3 · LT)
+ LP · (c2 + d2 · IT + e2 · LT) 
+ LP2 · (o3 + p3 · IT + q3 · LT) ,
+ LP2 · (f2 + g2 · IT + h2 · LT)
 (A.21)
+ LX · (i2 + j2 · IT + k2 · LT)  
τcor
+ LP · (l2 + m2 · IT + n2 · LT) lg = (a4 · IT + b4 · LT)
τref,4

+ LP2 · (o2 + p2 · IT + q2 · LT) + LP · (c4 + d4 · IT + e4 · LT)

+ LX2 · (r2 + s2 · IT + t2 · LT) + LP2 · (f4 + g4 · IT + h4 · LT)
B. Imbert et al. / Combustion and Flame 155 (2008) 380–408 407


+ LΦ · (i4 + j4 · IT + k4 · LT) References
+ LP · (l4 + m4 · IT + n4 · LT)
 [1] B. Imbert, L. Catoire, N. Chaumeix, C.E. Paillard, J.
+ LP2 · (o4 + p4 · IT + q4 · LT) Propul. Power 20 (2004) 415–426.

+ LΦ 2 · (r4 + s4 · IT + t4 · LT) [2] B. Imbert, L. Catoire, N. Chaumeix, C.E. Paillard, in:
Proceedings of the 1st European Combustion Meeting,
+ LP · (u4 + v4 · IT + w4 · LT) Orleans, France, 2003.
 [3] B. Imbert, L. Catoire, N. Chaumeix, G. Dupré, C.E.
+ LP2 · (x4 + y4 · IT + z4 · LT) .
Paillard, Proc. Combust. Inst. 30 (2004) 1925–1931.
(A.22) [4] B. Imbert, L. Catoire, N. Chaumeix, G. Dupré, C.E.
Correlations (A.21) and (A.22) provide the delays Paillard, in: Proceedings of the 2nd European Combus-
τ cor associated with n-C7 H16 /O2 mixtures, respec- tion Meeting, Louvain la Neuve, Belgium, 2005.
tively rich (Φ ∈ [(Φref = 1.0), 2.5]) and lean (Φ ∈ [5] D. Desbordes, C. Canteins, F. Franzetti, R. Zitoun, B.
[0.25, (Φref = 1.0)]). The relative deviation {τcor | Khasainov, in: Proceedings of the 19th International
Colloquium on the Dynamics of Explosions and Re-
τsim } remains lower than:
active Systems, Hakone, Japan, 2003, No. 33.
[6] H.J. Curran, P. Gaffuri, W.J. Pitz, C.K. Westbrook,
(1) 1.5% (0.3% on average) for the first-order poly-
Combust. Flame 114 (1998) 149–177.
nomial in LΦ, considering: [7] H.J. Curran, P. Gaffuri, W.J. Pitz, C.K. Westbrook,
• 48 time values τ sim a priori used to establish Combust. Flame 129 (2002) 253–280.
the database; [8] D.L. Chapman, Philos. Mag. 47 (1899) 90–104.
• 12 time values τ sim a posteriori evaluated for [9] E. Jouguet, J. Math. Pures Appl. (6) 60 (1905) 347–425
random conditions; (in French).
(2) 3.5% (0.6% on average) for the second-order [10] Ya.B. Zel’dovitch, Zh. Eksp. Teoret. Fiz. 10 (1940)
polynomial in LΦ, considering: 542–568 (in Russian); available translation: NACA
• 59 time values τ sim a priori used to establish TM 1261 (1950).
the database; [11] J. von Neumann, OSRD Report No. 549, USA, 1942.
• 18 time values τ sim a posteriori evaluated for [12] W. Döring, Ann. Phys. 43 (1943) 421–436 (in German).
random conditions. [13] D.F. Davidson, R.K. Hanson, Int. J. Chem. Kinet. 36
(2004) 510–523.
The coefficient analysis shows that: [14] S.A. Arrhenius, Z. Phys. Chem. 4 (1889) 226–248 (in
German).
(1) the difference {correlation τcor | reference τsim } [15] N.N. Semenov, Tsepnye Reaktsii, Goskhimizdat,
appears larger than in the previous cases, due to: Leningrad, 1934 (in Russian); available translation:
Chemical Kinetics and Chain Reactions, Oxford Univ.
• the complex evolution of τsim = f (Φ), not
Press, London, 1935.
easily reproducible with low-order polynomi-
[16] A. Lifshitz, K. Scheller, A. Burcat, G.B. Skinner, Com-
als; bust. Flame 16 (1971) 311–321.
• the additional need to check the quasi-tangency [17] Y. Auffret, D. Desbordes, H.N. Presles, Shock
of the correlations related to rich and quite lean Waves 11 (2001) 89–96.
mixtures (use of the stoichiometric composi- [18] C.K. Westbrook, Combust. Flame 46 (1982) 191–210.
tion as the connection zone); [19] Ya.B. Zeldovich, S.M. Kogarko, M.N. Simonov, Zh.
(2) the coefficients τ ref, 3 , a3 −h3 and τ ref, 4 , a4 −h4 Tekh. Fiz. 26 (1956) 1744–1768 (in Russian); avail-
remain comparable to their counterparts τ ref, 1 , able translation: Sov. Phys. Tech. Phys. 1 (1956) 1689–
a1 − h1 defined for the reference mixture, a con- 1713.
sequence (as in the diluted case) of the lost influ- [20] T. Turányi, Comput. Chem. 18 (1994) 45–54.
ence of the corrective terms i3 − q3 and i4 − z4 [21] T. Turányi, Proc. Combust. Inst. 25 (1994) 949–955.
(related to the powers of LΦ) for mixtures ap- [22] A.S. Tomlin, T. Turányi, M.J. Pilling, in: Low-Tempe-
proaching stoichiometry (Φ = 1 and LΦ = 0); rature Combustion and Autoignition, in: Comprehen-
(3) the impact of Φ is more marked on τ , than the sive Chemical Kinetics, vol. 35, Elsevier, Amsterdam,
X Ar one: 1997, Ch. 4.
• the absolute values of the high-order coeffi- [23] F.A. Lindemann, S. Arrhenius, I. Langmuir, N.R. Dhar,
J. Perrin, W.C. Lewis, Trans. Faraday Soc. 17 (1922)
cients i3 − q3 and i4 − z4 remain more impor-
598–606.
tant than their counterparts related to the argon
[24] J. Troe, J. Phys. Chem. 83 (1979) 114–126.
addition i2 − z2 ;
[25] C. Chevalier, P. Louessard, U.C. Müller, J. Warnatz,
• the basic terms a3 −h3 and a4 −h4 differ more in: Proceedings of the 2nd International Symposium on
from the reference mixture ones a1 − h1 than Diagnostics and Modeling of Combustion in Recipro-
from the diluted compositions ones a2 − h2 . cating Engines, Tokyo, Japan, 1990, pp. 93–97.
408 B. Imbert et al. / Combustion and Flame 155 (2008) 380–408

[26] D.J. Vermeer, J.W. Meyer, A.K. Oppenheim, Combust. [34] R. Brun, Introduction à la dynamique des gaz réactifs,
Flame 18 (1972) 327–336. Cépaduès, France, 2006 (in French).
[27] C.M. Coats, A. Williams, Proc. Combust. Inst. 17 [35] R. Penrose, Proc. Cambridge Philos. Soc. 51 (1955)
(1978) 611–621. 406–413.
[28] A. Burcat, R.F. Farmer, R.A. Matula, in: Proceedings [36] B. Taylor, Methodus incrementorum directa et inversa,
of the 13th International Symposium on Shock Tubes in: Proceedings of the Royal Society, London, 1715 (in
and Waves, Albany, NY, 1981, pp. 826–833. Latin).
[29] M.B. Colket, L.J. Spadaccini, J. Propul. Power 17 [37] I. Newton, Method of fluxions, in: Proceedings of the
(2001) 315–323. Royal Society, London, 1671.
[30] D.C. Horning, D.F. Davidson, R.K. Hanson, J. Propul. [38] J. Raphson, Analysis aequationum universalis, in: Pro-
Power 18 (2002) 363–371. ceedings of the Royal Society, London, 1690 (in Latin).
[31] J.M. Smith, J.M. Simmie, H.J. Curran, Int. J. Chem. [39] J.A. Nelder, R. Mead, Comput. J. 8 (1965) 308–313.
Kinet. 37 (2005) 728–736. [40] H.P. Schwefel, Evolution and Optimum Seeking, Sixth
[32] R.A. Marcus, J. Chem. Phys. 20 (1952) 359–364. Generation Computer Technologies Series, Research
[33] L. Catoire, Ph.D. thesis, University of Orléans, France, Studies Press, John Wiley & Sons, New York, 1995.
1995 (in French). [41] K. Pearson, Philos. Mag. 50 (1900) 157–175.

S-ar putea să vă placă și