A data-driven subgrid scale model in Large Eddy Simulation of turbulent premixed combustion

Shin, Junsu; Ge, Yuehe; Lampmann, Arne; Pfitzner, Michael

doi:10.1016/j.combustflame.2021.111486

Cited by 18 publications

(6 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…MDNs are machine learning models, which combine artificial neural networks with probabilistic mixture models to represent conditional densities (Bishop 2006). Their use has increased rapidly in recent years with applications in a variety of fields for a range of reduced order modelling and emulation tasks, including surrogate modelling of fluid flow (Maulik et al 2020), parameterisation of subgrid momentum forcing in ocean models (Guillaumin & Zanna 2021), emulation of complex stochastic models in epidemiology (Davis et al 2020) and multi-scale models of chemical reaction networks (Bortolussi & Palmieri 2018), and subgrid scale closures in large eddy simulations of turbulent combustion (Shin et al 2021).…”

Section: Introductionmentioning

confidence: 99%

Learning fluid dynamical statistics using stochastic neural networks

T.¹

2023

Preprint

View full text Add to dashboard Cite

<div> <div> <div>Many practical problems in fluid dynamics demand an empirical approach, where statistics estimated from data inform understanding and modelling. In this context data-driven probabilistic modelling offers an elegant alternative to ad hoc estimation procedures. Probabilistic models are useful as emulators, but also offer an attractive means of estimating particular statistics of interest. In this paradigm one can rely on proper scoring rules for model comparison and validation, and invoke Bayesian statistics to obtain rigorous uncertainty quantification. Stochastic neural networks provide a particularly rich class of probabilistic models, which, when paired with modern optimisation algorithms and GPUs, can be remarkably efficient. We demonstrate this approach by learning the single particle transition density of ocean surface drifters from decades of Global Drifter Program observations using a Bayesian mixture density network. From this we derive maps of various displacement statistics and corresponding uncertainty maps. Our model also offers a means of simulating drifter trajectories as a discrete-time Markov process, which could be used to study the transport of plankton or plastic in the upper ocean.</div> </div> </div>

show abstract

Section: Introductionmentioning

confidence: 99%

Learning fluid dynamical statistics using stochastic neural networks

T.¹

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…MDNs are machine learning models, which combine artificial neural networks with probabilistic mixture models to represent conditional densities (Bishop, 2006). Their use has increased rapidly in recent years with applications in a variety of fields for a range of reduced order modeling and emulation tasks, including surrogate modeling of fluid flow (Maulik et al., 2020), parameterization of subgrid momentum forcing in ocean models (Guillaumin & Zanna, 2021), emulation of complex stochastic models in epidemiology (C. N. Davis et al., 2020) and multi‐scale models of chemical reaction networks (Bortolussi & Palmieri, 2018), and subgrid scale closures in large eddy simulations of turbulent combustion (Shin et al., 2021).…”

Section: Introductionmentioning

confidence: 99%

Inferring Ocean Transport Statistics With Probabilistic Neural Networks

2023

J Adv Model Earth Syst

View full text Add to dashboard Cite

Using a probabilistic neural network and Lagrangian observations from the Global Drifter Program, we model the single particle transition probability density function (pdf) of ocean surface drifters. The transition pdf is represented by a Gaussian mixture whose parameters (weights, means, and covariances) are continuous functions of latitude and longitude determined to maximize the likelihood of observed drifter trajectories. This provides a comprehensive description of drifter dynamics allowing for the simulation of drifter trajectories and the estimation of a wealth of dynamical statistics without the need to revisit the raw data. As examples, we compute global estimates of mean displacements over 4 days and lateral diffusivity. We use a probabilistic scoring rule to compare our model to commonly used transition matrix models. Our model outperforms others globally and in three specific regions. A drifter release experiment simulated using our model shows the emergence of concentrated clusters in the subtropical gyres, in agreement with previous studies on the formation of garbage patches. An advantage of the neural network model is that it provides a continuous‐in‐space representation and avoids the need to discretize space, overcoming the challenges of dealing with nonuniform data. Our approach, which embraces data‐driven probabilistic modeling, is applicable to many other problems in fluid dynamics and oceanography.

show abstract

“…To tackle this issue, they introduce the so-called residual regression model by replacing convolution and pooling layers into fully connected (FC) layers in ResNet. By maintaining the shortcut within residual blocks, residual regression enhances data flow in the neural network and has been applied in many fields, such as computational fluid dynamics [ 30 , 31 ], computer-aided geometric design [ 32 ], and safety control in visual serving applications [ 33 , 34 ]. Besides the localness of convolution, the independence of input features also requires the replacement of convolution layers in a neural network regression model.…”

Section: Introductionmentioning

confidence: 99%

Densely Connected Neural Networks for Nonlinear Regression

Jiang

Chen

et al. 2022

Entropy

View full text Add to dashboard Cite

Densely connected convolutional networks (DenseNet) behave well in image processing. However, for regression tasks, convolutional DenseNet may lose essential information from independent input features. To tackle this issue, we propose a novel DenseNet regression model where convolution and pooling layers are replaced by fully connected layers and the original concatenation shortcuts are maintained to reuse the feature. To investigate the effects of depth and input dimensions of the proposed model, careful validations are performed by extensive numerical simulation. The results give an optimal depth (19) and recommend a limited input dimension (under 200). Furthermore, compared with the baseline models, including support vector regression, decision tree regression, and residual regression, our proposed model with the optimal depth performs best. Ultimately, DenseNet regression is applied to predict relative humidity, and the outcome shows a high correlation with observations, which indicates that our model could advance environmental data science.

show abstract

A data-driven subgrid scale model in Large Eddy Simulation of turbulent premixed combustion

Cited by 18 publications

References 38 publications

Learning fluid dynamical statistics using stochastic neural networks

Learning fluid dynamical statistics using stochastic neural networks

Inferring Ocean Transport Statistics With Probabilistic Neural Networks

Densely Connected Neural Networks for Nonlinear Regression

Contact Info

Product

Resources

About