Reduced order modeling of fluid flows using convolutional neural networks

Fukagata, Koji

doi:10.1299/jfst.2023jfst0002

Cited by 4 publications

(1 citation statement)

References 54 publications

(61 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In other words, understanding the nature of turbulent mass transfers and predicting their states and behaviors have been based on statistical features, but now is the time to discuss the possibility of extracting significant features from the instantaneous fields with the help of machine/deep learning. Recently advanced deep learning has been utilized to infer from a large number of instantaneous turbulence snapshots, the so-called turbulence big data (Fukagata and Fukami, 2020;Fukagata, 2023), and to realize a superresolution reconstruction (Fukami and Fukagata, 2019), a turbulence quantification (Corbetta et al, 2021), a state estimation (Wang and Zaki, 2021), and an optimal control (Yugeta et al, 2023).…”

Section: Introductionmentioning

confidence: 99%

Deep learning estimation of scalar source distance for different turbulent and molecular diffusion environments

TSUKAHARA,

ISHIGAMI,

IRIKURA

2024

JFST

View full text Add to dashboard Cite

In order to adopt convolutional neural networks (CNN) for practical use in estimating the source of scalar dispersion in turbulent flows, such as gas leaks in industrial plants, the inference accuracy was verified using scalar concentration distributions in various turbulent environmental conditions. Training and test data were obtained through quasi direct numerical simulations on a flow system with a scalar-source-attached cylinder downstream of a turbulent grid, at two Schmidt numbers. An inference accuracy of at least 90% was confirmed under trained flow conditions, provided that the observation window was large enough to capture the integral scale in scalar fluctuations in turbulence. When the flow conditions (in terms of the turbulence intensity and the Schmidt number) differed between the training and testing phases, the accuracy was significantly reduced. However, the learner supervised with images under all different conditions showed high generalization performance. Singular value decomposition was used to discuss the features essential for training and prediction. In our CNN inference, we concluded that the source estimation was achieved by extracting from the integral-scale scalar patch distribution the features related to each of the turbulent and molecular diffusion progressions and their correlations. The results provide insights into the potential solutions for accurately predicting the sources of material dispersion in turbulent environments.

show abstract

Section: Introductionmentioning

confidence: 99%

Deep learning estimation of scalar source distance for different turbulent and molecular diffusion environments

TSUKAHARA,

ISHIGAMI,

IRIKURA

2024

JFST

View full text Add to dashboard Cite

show abstract

CNN for scalar-source distance estimation in grid-generated turbulence

Someya,

Araki,

Tsukahara

2025

Applied Thermal Engineering

View full text Add to dashboard Cite

Flow control by a hybrid use of machine learning and control theory

Ishize,

Omichi,

Fukagata

2024

HFF

View full text Add to dashboard Cite

Purpose Flow control has a great potential to contribute to a sustainable society through mitigation of environmental burden. However, the high dimensional and nonlinear nature of fluid flows poses challenges in designing efficient control laws using the control theory. This paper aims to propose a hybrid method (i.e. machine learning and control theory) for feedback control of fluid flows, by which the flow is mapped to the latent space in such a way that the linear control theory can be applied therein. Design/methodology/approach The authors propose a partially nonlinear linear system extraction autoencoder (pn-LEAE), which consists of convolutional neural networks-based autoencoder (CNN-AE) and a custom layer to extract low-dimensional latent dynamics from fluid velocity field data. This pn-LEAE is designed to extract a linear dynamical system so that the modern control theory can easily be applied, while a nonlinear compression is done with the autoencoder (AE) part so that the latent dynamics conform to that linear system. The key technique is to train this pn-LEAE with the ground truths at two consecutive time instants, whereby the AE part retains its capability as the AE, and the weights in the linear dynamical system are trained simultaneously. Findings The authors demonstrate the effectiveness of the linear system extracted by the pn-LEAE, as well as the designed control law’s effectiveness for a flow around a circular cylinder at the Reynolds number of ReD = 100. When the control law derived in the latent space was applied to the direct numerical simulation, the lift fluctuations were suppressed over 50%. Originality/value To the best of the authors’ knowledge, this is the first attempt using CNN-AE for linearization of fluid flows involving transient development to design a feedback control law.

show abstract

Reduced order modeling of fluid flows using convolutional neural networks

Cited by 4 publications

References 54 publications

Deep learning estimation of scalar source distance for different turbulent and molecular diffusion environments

Deep learning estimation of scalar source distance for different turbulent and molecular diffusion environments

CNN for scalar-source distance estimation in grid-generated turbulence

Flow control by a hybrid use of machine learning and control theory

Contact Info

Product

Resources

About