Identifying key differences between linear stochastic estimation and neural networks for fluid flow regressions

Nakamura, Taichi; Fukami, Kai; Fukagata, Koji

doi:10.1038/s41598-022-07515-7

Cited by 21 publications

(4 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Preparing high-quality input data is essential for successfully reconstructing turbulent flows. However, it is necessary to assess the robustness and sensitivity of the models against noisy inputs [182]. This point will be important as machine-learning-based super-resolution analyses become utilized in industrial applications [183].…”

Section: Discussionmentioning

confidence: 99%

Super-resolution analysis via machine learning: a survey for fluid flows

Fukami

Fukagata

Taira

2023

Theor. Comput. Fluid Dyn.

Self Cite

View full text Add to dashboard Cite

This paper surveys machine-learning-based super-resolution reconstruction for vortical flows. Super resolution aims to find the high-resolution flow fields from low-resolution data and is generally an approach used in image reconstruction. In addition to surveying a variety of recent super-resolution applications, we provide case studies of super-resolution analysis for an example of two-dimensional decaying isotropic turbulence. We demonstrate that physics-inspired model designs enable successful reconstruction of vortical flows from spatially limited measurements. We also discuss the challenges and outlooks of machine-learning-based super-resolution analysis for fluid flow applications. The insights gained from this study can be leveraged for super-resolution analysis of numerical and experimental flow data. Graphical abstract

show abstract

Section: Discussionmentioning

confidence: 99%

Super-resolution analysis via machine learning: a survey for fluid flows

Fukami

Fukagata

Taira

2023

Theor. Comput. Fluid Dyn.

Self Cite

View full text Add to dashboard Cite

show abstract

“…For instance, consider the nonlinear wavelet decomposition (see e.g. Farge & Schneider, 2006), or more recent data-driven methods, such as convolutional neural networks (CNNs) POD (see Guastoni et al, 2021;Güemes, Discetti, & Ianiro, 2019;Nakamura, Fukami, & Fukagata, 2022). A CNN POD method considers the same spatial basis as traditional POD; however, the time coefficients are estimated using a CNN, making it able to account for nonlinear features in the flow.…”

Section: Discussionmentioning

confidence: 99%

Proper orthogonal decomposition modal analysis in a baffled stirred tank: a base tool for the study of structures

Arosemena,

Solsvik

2023

Flow

View full text Add to dashboard Cite

Proper orthogonal decomposition (POD) is applied to three-dimensional (3-D) velocity fields collected from large-eddy simulations (LES) of a baffled stirred tank. In the LES, the tank operates with a Rushton-type impeller under turbulent conditions (at least in the near-impeller region) and the working fluid exhibits either Newtonian or shear-thinning rheology. The most energetic POD modes are analysed, and a POD reconstruction based on the higher modes is proposed to approximate the fluctuating component of the velocity field. Subsequently, the POD reconstruction is used to identify vortical structures and characterise them in terms of their shape. The structures are identified by considering a frame-invariant formulation of a popular, Eulerian, local-region-type method: the $Q$ -criterion. Statistics of shape-related parameters are then investigated to address the morphology of the structures. It is found that: (i) regardless of the working fluid rheology, it seems feasible to decompose the 3-D field into its mean, most energetic periodic and fluctuating components using POD, allowing, for instance, reduced-order modelling of the energetic periodic motions for mixing enhancement purposes, and (ii) vortical structures related to turbulence are mostly tubular. Finding (ii) implies that, as starting point, phenomenological models for the interaction between fluid particles (drops and bubbles) and vortices should consider the latter as cylindrical structures rather than of spherical shape, as classically assumed in these models.

show abstract

“…In general, SVD can be performed on large data sets of X = [x 1 x 2 • • • x m ] ∈ C n×m to extract dominant patterns from low-dimensional approximations to high dimensions for that data set (Brunton and Kuts, 2022;Nakamura et al, 2022). Eigenvectors can be extracted from m sets of measured values or image data as the row components, where the column components are individual measured values or image data converted to a one-dimensional array as x ∈ C n .…”

Section: Inference Accuracy For Low-dimensional Svd Modesmentioning

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

Identifying key differences between linear stochastic estimation and neural networks for fluid flow regressions

Cited by 21 publications

References 34 publications

Super-resolution analysis via machine learning: a survey for fluid flows

Super-resolution analysis via machine learning: a survey for fluid flows

Proper orthogonal decomposition modal analysis in a baffled stirred tank: a base tool for the study of structures

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

Contact Info

Product

Resources

About