Three-dimensional direct simulations and structure of expanding turbulent methane flames
Introduction
Localized ignition of a reactive mixture and later flame development in a turbulent flow represents a very important combustion process, found in many practical applications. Current understanding is still unsatisfactory and severely limits prediction capabilities, and thus achievements concerning the control of pollutant emissions or improvement of the global efficiency.
Direct numerical simulations (DNS) have been known for over 10 years to be ideally suited to study turbulent premixed flames [1], [2], [3] because they do not require any particular assumption concerning the turbulence. Due to the huge cost of direct simulations, strong hypotheses have generally been introduced to reduce the requested computing times [4]. Despite these simplifications, like, for example, constant-density combustion [5] or single-step chemistry [1], [6], [7], many interesting results have been obtained over the years.
When considering quantitative problems like predictions of intermediate radicals, pollutant emission or ignition/extinction limits, the reaction processes should be described using detailed models [2], [8], [9], [10], [11], [12]. This leads to a more faithful representation of the combustion processes, as explained in a recent review paper [13].
Very often, such computations relying on detailed models have been limited to 2D. Three-dimensional direct simulations including detailed reaction schemes are fairly recent. Associated computing times are huge and limit the possible use of such simulations to investigate turbulent combustion. Nevertheless, several such publications have appeared in the last three years concerning premixed as well as non-premixed combustion [14], [15], [16], [17], [18], [19], [20], [21], [22].
The present work builds on top of our recent efforts to carry out three-dimensional direct simulations of reacting flows at reasonable computing costs, while keeping a realistic description of chemical processes. Such computations are clearly required, since: (1) turbulence is intrinsically three-dimensional, and the behavior of 2D-turbulence is very different; (2) reducing chemical processes to a single-step reaction is a crude approximation, very useful to get global trends but often too limited when quantitative results are needed, as shown in [13].
Concerning previous simulations on the same configuration, two-dimensional DNS of spark ignition in a turbulent flow using single-step chemistry are reported in [4], [9], [23], where the emphasis is laid on flame initiation and on modeling issues. Two-dimensional direct simulations using detailed chemistry are presented in [11]. High-resolution three-dimensional simulations relying on single-step chemistry have also been presented at the last Symposium [7], where the emphasis was laid on flame curvature.
In the last Symposium volume, we published a study dealing with the comparison of such two-dimensional and three-dimensional direct simulations, showing large discrepancies between both [16]. A three-dimensional DNS code, called π3 and written in Fortran 95, has been specifically developed for this purpose. This same code is again used in the present study.
Physical models and numerical methods employed in π3 are first briefly recalled. Afterwards, results of three-dimensional direct simulations are described. We first analyze our results in the light of turbulent combustion regimes. Many quantities, in particular flame surface area, stretch-rate, flame front curvature, and flame thickness, are then extracted from numerical data. For flame curvature, we compare the results obtained using a full three-dimensional treatment with those of post-processing methods working on 2D slices. Impact of turbulence on flame thickness is finally investigated, partly supporting some recent findings [24].
Section snippets
Physical models and numerical methods
Our latest DNS code, called π3, has been already described in previous publications, e.g. [16], [22]. We therefore only consider briefly two specific aspects of this code.
First, our code relies on a low-Mach number approximation and not on a fully compressible formulation. For most applications of interest, the maximum Mach number is generally quite small, at least in the region where combustion takes place. It is then unnecessary and computationally inefficient to employ a fully compressible
Direct simulations and results
Fully premixed methane/air flames at an equivalence ratio of ϕ = 1.59, atmospheric pressure, and fresh gas temperature of 298 K are considered. This high value of ϕ has been retained to facilitate the computations, and in particular to get a larger flame thickness and lower density jump through the flame, leading to a smoother time- and space-integration. Moreover, previous studies tend to show that differences with single-step chemistry are considerably larger for rich flames [8], [13]. These
Conclusions
We have successfully developed a three-dimensional direct simulation code relying on the low-Mach number approximation and on the FPI chemistry reduction technique. The obtained tool is computationally very efficient and allows three-dimensional DNS of turbulent flames at moderate Reynolds numbers on a standard PC. This code has been used to investigate the development of a premixed methane/air flame in a turbulent flow, as observed after ignition. Evolutions of the flame surface area, of the
Acknowledgments
Many people have participated in the development of the code π3. We in particular thank O. Gicquel, J. de Charentenay, and R. Hilbert for their contributions.
References (36)
Proc. Combust. Inst.
(1996)- et al.
Combust. Flame
(1993) - et al.
Proc. Combust. Inst.
(2002) - et al.
Combust. Flame
(2000) - et al.
Proc. Combust. Inst.
(2000) - et al.
Combust. Flame
(2002) - et al.
Prog. Energy Combust. Sci.
(2004) - et al.
Proc. Combust. Inst.
(2000) - et al.
Proc. Combust. Inst.
(2002) - et al.
Proc. Combust. Inst.
(2002)
Proc. Combust. Inst.
Proc. Combust. Inst.
Proc. Combust. Inst.
Combust. Flame
Proc. Combust. Inst.
Proc. Combust. Inst.
Cited by (60)
Combustion machine learning: Principles, progress and prospects: Combustion machine learning
2022, Progress in Energy and Combustion ScienceMorphology and structure of spherically propagating premixed turbulent hydrogen - air flames
2022, Combustion and FlameCitation Excerpt :However, as the mentioned methodologies focus on the overall flame surface properties, the discussion related to the development of local flame structures would be limited. On the other hand, with DNS, it is possible to investigate local turbulence-flame interactions in developing-flame fronts, such as the local variations in burning rates and turbulent stretching effects, as well as the global evolution of the flame surface and brush thickness [17–24,77]. However, in many studies, to reduce the huge cost of the DNS, (i) single-step chemistry or reduced reaction mechanisms were used instead of detailed chemistry, (ii) single/fixed Lewis number assumption was used instead of detailed diffusion, and/or (iii) resolution was reduced.
A direct numerical simulation of Jet A flame kernel quenching
2021, Combustion and FlameCitation Excerpt :Thevenin et al. compared 2D and 3D results for a weakly turbulent flame kernel, which showed that the 2D simulations could not accurately capture the flame stretch or kernel growth rate [6]. A subsequent study quantified the distribution of flame curvatures with respect to Reynolds number (Re), which showed that fluctuations in curvature increased with Re while the mean value remains nearly unchanged [7]. Thevenin et al. have also highlighted the need to perform multiple simulations in order to capture the statistics of natural variations in the initial turbulent field [7,15].
DNS study of the global heat release rate during early flame kernel development under engine conditions
2020, Combustion and FlameCitation Excerpt :With the occurrence of the plateau, the total area rate-of-change of Engine Kernel I becomes larger than in Engine Kernel II, which reduces the difference in flame area between both kernel realizations (cf. Fig. 3 (b)). Similar deviations in global stretch rate were previously observed between flame kernel realizations in two- [56] and three-dimensional [57] DNS. It will be shown in Section 4.2.2 that also in local regions of the planar flame, net area rate-of-change fluctuates noticeably, even after the flame is on average fully developed.
Evolution of averaged local premixed flame thickness in a turbulent flow
2019, Combustion and FlameCitation Excerpt :While such thickness variations have been attracting significant attention, this fundamental issue is still intricate and the available data are controversial. Indeed, on the one hand there are experimental [1,2] and Direct Numerical Simulation (DNS) data [3] that indicate local thinning of flames in turbulent flows. On the other hand, there are experimental [4–6] and DNS data [7–12] that show the opposite trend.