This work presents a new approach for premixed turbulent combustion modeling based on convolutional neural networks (CNN). We first propose a framework to reformulate the problem of subgrid flame surface density estimation as a machine learning task.Data needed to train the CNN is produced by direct numerical simulations (DNS) of a premixed turbulent flame stabilized in a slot-burner configuration. A CNN inspired from a U-Net architecture is designed and trained on the DNS fields to estimate sub-grid scale wrinkling. It is then tested on an unsteady turbulent flame where the mean inlet velocity is increased for a short time and the flame must react to a varying turbulent incoming flow. The CNN is found to efficiently extract the topological nature of the flame and predict subgrid scale wrinkling, outperforming classical algebraic models.This method can be seen as a data-driven extension of dynamic formulations, where topological information was extracted in a hand-designed fashion.
Using homogeneous lean mixtures is an efficient way to reduce fuel consumption and pollutant emissions in internal combustion engines. However, lean combustion requires breakthrough technologies to induce reliable ignition and fast combustion. One of these technologies uses pre-chamber to create multiple hot turbulent jets and provide ignition sites for the lean mixture. In this paper, the behaviour of a pre-chamber ignition system used to ignite the main chamber of a real engine is studied using large eddy simulation with direct integration of analytically reduced chemistry using the dynamic thickened flame model. The large eddy simulation allows to analyze the flow entering and leaving the prechamber, to measure the cooling and quenching effects introduced by the hot gas passages through the ducts connecting pre-and main chambers and to analyze the ignition and combustion sequences. For the case studied here, small amount of flame kernels are exhausted from the pre-chamber. Hot products penetrate the main chamber, disperse and mix with the fresh reactants and lead to ignition. The combustion in the main chamber starts in a distributed reaction mode before reaching a flamelet propagation mode.
The Poisson equation is present in very different domains of physics and engineering. In most cases, this equation can not be solved directly and iterative solvers are used. For many solvers, this step is computationally intensive. In this study, an alternative resolution method based on neural networks is evaluated for incompressible flows. A fluid solver coupled with a Convolutional Neural Network is developed and trained on random cases with constant density to predict the pressure field. Its performance is tested in a plume configuration, with different buoyancy forces, parametrized by the Richardson number. The neural network is compared to a traditional Jacobi solver. The performance improvement is considerable, although the accuracy of the network is found to depend on the flow operating point: low errors are obtained at low Richardson numbers, whereas the fluid solver becomes unstable with large errors for large Richardson number. Finally, a hybrid strategy is proposed in order to benefit from the calculation acceleration while ensuring a user-defined accuracy level. In particular, this hybrid CFD-NN strategy, by maintaining the desired accuracy whatever the flow condition, makes the code stable and reliable even at large Richardson numbers for which the network was not trained for. This study demonstrates the capability of the hybrid approach to tackle new flow physics, unseen during the network training.
International audienceThe prediction of Autoignition (AI) delay is an essential prerequisite to account for abnormal combustions (e.g. knock or super knock) that can appear in Internal Combustion (IC) engines. In this paper, a simple model called Ignition to Propagation Reduced Scheme (IPRS) is proposed to add AI predictions in reduced chemical schemes, which are classically used to compute in-cylinder combustion in the context of Large Eddy Simulations (LES). The IPRS principle is to use a single two-reaction reduced scheme and adapt the pre-exponential factor of the fuel oxidation reaction as a function of the temperature: one value is used at low temperatures to correctly predict AI delays and an other one can be used at higher temperatures, where heat release occurs, to keep the flame propagation properties of the chemical scheme. After a first section that introduces the model, Perfectly Stirred Reactors and 1D flames simulations are used to verify that: (1) the modification of the pre-exponential constant of the Arrhenius law at low temperature does not alter the propagation properties of the reduced scheme and (2) this modification is sufficient to accurately predict AI delays. In a following section this model is implemented in a 3D LES solver to compute AI in a simplified IC engine for which results with complex chemistries are available and may be used as a reference for comparison. In the last section this model is applied to a highly downsized engine to illustrate its ability to predict AI in real engine configurations
International audienceLarge Eddy Simulation of knocking in piston engines requires high-fidelity physical models and numerical techniques. The need to capture temperature fields with high precision to predict autoignition is an additional critical constraint compared to existing LES in engines. The present work presents advances for LES of knocking in two fields: (1) a Conjugate Heat Transfer (CHT) technique is implemented to compute the flow within the engine over successive cycles with LES together with the temperature field within the cylinder head walls and the valves and (2) a reduced two-step scheme is used to predict both propagating premixed flames as well as autoignition times over a wide range of equivalence ratios, pressures and temperatures. The paper focuses on CHT which is critical for knocking because the gas temperature field is controlled by the wall temperature field and knocking is sensitive to small temperature changes. The CHT LES is compared to classical LES where the temperatures of the head and the valves are supposed to be homogeneous and imposed empirically. Results show that the skin temperature field (which is a result of the CHT LES while it is a user input for classical LES) is complex and controls knocking events. While the results of the CHT LES are obviously better because they suppress a large part of the empirical specification of the wall temperatures, this study also reveals a difficult and crucial element of the CHT approach: the description of exhaust valves cooling which are in contact with the engine head for part of the cycle and not in the rest of the cycle, leading to difficulties for heat transfer descriptions between valves and head. The CHT method is successfully applied to an engine studied at IFP Energies Nouvelles where knocking characteristics have been studied over a wide range of conditions
and available online here Cet article fait partie du dossier thématique ci-dessous publié dans la revue OGST, Vol. 69, n°1, et téléchargeable ici
D o s s i e rDOSSIER Edited by/Sous la direction de : C. Angelberger
IFP Energies nouvelles International Conference / Les Rencontres Scientifiques d'IFP Energies nouvelles LES4ICE 2012 -Large Eddy Simulation for Internal Combustion Engine FlowsLES4ICE 2012 -La simulation aux grandes échelles pour les écoulements dans les moteurs à combustion interne
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.