Training convolutional neural networks to estimate turbulent sub-grid scale reaction rates

Lapeyre, Corentin; Misdariis, Antony; Cazard, Nicolas; Veynante, Denis; Poinsot, Thiérry

doi:10.1016/j.combustflame.2019.02.019

Cited by 142 publications

(68 citation statements)

References 46 publications

(55 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As discussed in the recent review by Duraisamy et al, 2 there have been several studies aimed at using machine learning to model turbulent flows, especially in the context of Reynolds-averaged Navier-Stokes (RANS) simulations. 3,4 Other applications to modelling the near-wall region of turbulent flows were reported by Milano and Koumoutsakos, 5 while Lapeyre et al 6 and Beck et al 7 have documented the possibility of using machine learning in designing subgrid-scale (SGS) models for large-eddy simulation (LES). Machine learning has also been applied to flow control, 8,9 development of low-dimensional models, 10 generation of inflow conditions 11 or structure identification in two-dimensional decaying turbulence, 12 and as discussed by Kutz 13 DNNs will progressively be more widely used in fluid mechanics in the coming years.…”

Section: Introductionmentioning

confidence: 99%

Predictions of turbulent shear flows using deep neural networks

et al. 2019

View full text Add to dashboard Cite

In the present work we assess the capabilities of neural networks to predict temporally evolving turbulent flows. In particular, we use the nine-equation shear flow model by Moehlis et al. [New J. Phys. 6, 56 (2004)] to generate training data for two types of neural networks: the multilayer perceptron (MLP) and the long short-term memory (LSTM) network. We tested a number of neural network architectures by varying the number of layers, number of units per layer, dimension of the input, weight initialization and activation functions in order to obtain the best configurations for flow prediction. Due to its ability to exploit the sequential nature of the data, the LSTM network outperformed the MLP. The LSTM led to excellent predictions of turbulence statistics (with relative errors of 0.45% and 2.49% in mean and fluctuating quantities, respectively) and of the dynamical behavior of the system (characterized by Poincaré maps and Lyapunov exponents). This is an exploratory study where we consider a low-order representation of near-wall turbulence. Based on the present results, the proposed machine-learning framework may underpin future applications aimed at developing accurate and efficient data-driven subgrid-scale models for large-eddy simulations of more complex wall-bounded turbulent flows, including channels and developing boundary layers.

show abstract

Section: Introductionmentioning

confidence: 99%

Predictions of turbulent shear flows using deep neural networks

et al. 2019

View full text Add to dashboard Cite

show abstract

“…Artificial neural networks (ANNs) were found accurate, and both CPU and memory efficient [18,19]. Along similar lines, convolutional neural networks (CNNs), which were originally developed for analyzing visual representations [20], have been introduced in turbulent combustion modeling for direct deconvolution of the filtered progress variable [21], for modeling the unresolved flame surface wrinkling [22] and for the direct reconstruction of unresolved sources and fluxes from mesh-resolved quantities in large-eddy simulation (LES) [23]. Compared to ANNs, CNNs reduce the number of connections per layer to stack more of them efficiently, thus increasing the depth of the neural network.…”

Section: Introductionmentioning

confidence: 99%

Machine learning for detailed chemistry reduction in DNS of a syngas turbulent oxy-flame with side-wall effects

Wan

Barnaud

Domingo

2021

Proceedings of the Combustion Institute

View full text Add to dashboard Cite

A novel chemistry reduction strategy based on convolutional neural networks (CNNs) is developed and applied to direct numerical simulation (DNS) of a turbulent non-premixed flame interacting with a cooled wall. The fuel syngas mixture is burning in pure oxygen. The training and the subsequent application of the CNN rely on the processing of two-dimensional (2D) images built from species mass fractions and temperature (CNN input), to predict the corresponding chemical sources at the center of the image (CNN output). This image-type treatment of chemistry is found to efficiently capture intermediate radicals species highly sensitive to the local flame topology. To reduce the CPU cost, a simplified 2D DNS database with detailed chemistry serves as reference and is used for training and testing the neural network. Comparisons are also made a posteriori against the same 2D DNS with a reduced chemical scheme specialized for syngas. Then, three-dimensional (3D) DNS are conducted either with CNN or the reduced chemistry for more a posteriori tests. The CNN reduced chemistry outperforms the reduced Arrhenius based mechanism in the prediction of radical species, such as monoatomic hydrogen, and also in terms of CPU cost.

show abstract

“…Data driven approaches have also been coupled with principal component analysis for developing closure models in turbulent combustion using experimental multi-scalar measurements [29] and for building digital-twins to progress towards furnace control [30]. Along similar lines, convolutional neural networks (CNN), based on image-like treatment of the thermochemical fields, were shown useful to tackle the modeling of turbulent flames [31][32][33] and their control [34].…”

Section: Introductionmentioning

confidence: 99%

Chemistry reduction using machine learning trained from non-premixed micro-mixing modeling: Application to DNS of a syngas turbulent oxy-flame with side-wall effects

Wan

Barnaud

Vervisch

et al. 2020

Combustion and Flame

View full text Add to dashboard Cite

A chemistry reduction approach based on machine learning is proposed and applied to direct numerical simulation (DNS) of a turbulent non-premixed syngas oxy-flame interacting with a cooled wall. The training and the subsequent application of artificial neural networks (ANNs) rely on the processing of 'thermochemical vectors' composed of species mass fractions and temperature (ANN input), to predict the corresponding chemical sources (ANN output). The training of the ANN is performed aside from any flow simulation, using a turbulent non-adiabatic non-premixed micro-mixing based canonical problem with a reference detailed chemistry. Heat-loss effects are thus included in the ANN training. The performance of the ANN chemistry is then tested aposteriori in a two-dimensional DNS against the detailed mechanism and a reduced mechanism specifically developed for the operating conditions considered. Then, three-dimensional DNS are performed either with the ANN or the reduced chemistry for additional a-posteriori tests. The ANN reduced chemistry achieves good agreement with the Arrhenius-based detailed and reduced mechanisms, while being in terms of CPU cost 25 times faster than the detailed mechanism and 3 times faster than the reduced mechanism when coupled with DNS. The major potential of the method lies both in its data driven character and in the handling of the stiff chemical sources. The former allows for easy implementation in the context of automated generation of case-specific reduced chemistry. The latter avoids the Arrhenius rates calculation and also the direct integration of stiff chemistry, both leading to a significant CPU time reduction.

show abstract

Training convolutional neural networks to estimate turbulent sub-grid scale reaction rates

Cited by 142 publications

References 46 publications

Predictions of turbulent shear flows using deep neural networks

Predictions of turbulent shear flows using deep neural networks

Machine learning for detailed chemistry reduction in DNS of a syngas turbulent oxy-flame with side-wall effects

Chemistry reduction using machine learning trained from non-premixed micro-mixing modeling: Application to DNS of a syngas turbulent oxy-flame with side-wall effects

Contact Info

Product

Resources

About