On the choice of closure complexity in anisotropic drag closures for filtered Two Fluid Models

Cloete, Jan Hendrik; Cloete, Schalk; Radl, Stefan; Amini, Shahriar

doi:10.1016/j.ces.2019.06.006

Cited by 41 publications

(32 citation statements)

References 36 publications

(111 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Further inclusion of the additional feature inputs only increases the complexity of trained models and is of little impact on improving model performance. The analogous conclusion has also been drawn very recently by the conventional SGM study of Cloete et al 14 Therefore, three‐marker choice will be adopted for the following application to integrate the ML with the coarse‐grid CFD model.…”

Section: Filtered Subgrid Model Developmentsupporting

confidence: 55%

“…Alternatively, subgrid models (SGMs) for interphase closure corrections such as the energy minimization multi‐scale (EMMS), 7‐9 filtered TFM (fTFM), 2,10‐15 spatially‐averaged TFM (SA‐TFM), 16‐18 and gradient‐based method, 19‐21 have been developed to account for effects of unresolved inhomogeneous structures. SGM enables a dramatic reduction of computational overhead by employing coarse grids while still maintaining high accuracy.…”

Section: Introductionmentioning

confidence: 99%

“…Unlike the above studies, Schneiderbauer 16‐18 followed the concepts of large eddy simulations 29 and developed a third‐class of SA‐TFM that models the cluster‐induced turbulence to formulate the subgrid drag correction using the scalar variance of the solids hold‐up and large‐scale flow fluctuations of two phases as markers. Recently, Cloete et al 14,30,31 presented the importance to include the physical anisotropic impacts in fTFM development, and they unveiled minor influences of increasing the SGM complexity via introducing additional markers. For more elaborate reviews, interested readers are referred to excellent articles of Schneiderbauer et al 32 and Sundaresan et al 4 Yet, despite multifarious contributions, better understanding subgrid closure modeling related to inhomogeneous structures and accurately predicting gas–particle flow hydrodynamics still pose prodigious challenges.…”

Section: Introductionmentioning

confidence: 99%

“…For more elaborate reviews, interested readers are referred to excellent articles of Schneiderbauer et al 32 and Sundaresan et al 4 Yet, despite multifarious contributions, better understanding subgrid closure modeling related to inhomogeneous structures and accurately predicting gas–particle flow hydrodynamics still pose prodigious challenges. For example, Cloete et al 14 indicated that a functional fitting method is commonly adopted in conventional filtered correlations; however, various model coefficients might have to be adjusted in such an approach. In this case, a novel predictive modeling approach for better representation of such filtered relations is needed.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Machine learning to assist filtered two‐fluid model development for dense gas–particle flows

2020

View full text Add to dashboard Cite

Machine learning (ML) is experiencing an immensely fascinating resurgence in a wide variety of fields. However, applying such powerful ML to construct subgrid interphase closures has been rarely reported. To this end, we develop two data-driven ML strategies (i.e., artificial neural networks and eXtreme gradient boosting) to accurately predict filtered subgrid drag corrections using big data from highly resolved simulations of gas-particle fluidization. Quantitative assessments of effects of various subgrid input markers on training prediction outputs are performed and three-marker choice is demonstrated to be the optimal one for predicting the unseen test set. We then develop a parallel data loader to integrate this predictive ML model into a computational fluid dynamic (CFD) framework. Subsequent coarse-grid simulations agree fairly well with experiments regarding the underlying hydrodynamics in bubbling and turbulent fluidized beds. The present ML approach provides easily extended ways to facilitate the development of predictive models for multiphase flows. K E Y W O R D S coarse-grid simulation, filtered two-fluid model, fine-grid simulation, fluidization, machine learning, subgrid closure model 1 | INTRODUCTION Gas-solid fluidized bed reactors (FBRs) are ubiquitous in process engineering. Unfortunately, inhomogeneous mesoscale flow structures in such reactors are incredibly complex and manifest persistent instabilities in flow properties spanning all sorts of spatiotemporal scales. 1-3 This dynamic nature is of fundamental importance in essential process features such as the interphase momentum, species, and heat transfer. 4 Development of accurate and reliable modeling methods for capturing such complex dynamics has been the grand challenge for FBR modelers. One of the broadly-practiced predictive methods is the kinetic theory of granular flow (KTGF) based two-fluid model (TFM), which treats the fluid and particle phases as interpenetrating continuums. 5 However, the TFM simulation with sufficiently fine grids is necessitated for resolving all relevant scales of flow hydrodynamics. Fullmer and Hrenya 3 indicated that a typical mesh spacing of~3.3d s is required to obtain the grid independency for simulation statistics at intermediate solid phase fraction, and the grid size less than 10d s for dilute clustering flows is demanded for computational accuracy. For dense bubbling FBRs, Wang et al. 6 recommended a grid resolution of 2-4 d s . Such high requirements (10 −5 -10 −3 m) for reactor-scale simulations (10 −1 -10 1 m) are computationally prohibitive.Alternatively, subgrid models (SGMs) for interphase closure corrections such as the energy minimization multi-scale (EMMS), [7][8][9] filtered TFM (fTFM), 2,10-15 spatially-averaged TFM (SA-TFM), 16-18 and gradient-based method, [19][20][21] have been developed to account for effects of unresolved inhomogeneous structures. SGM enables a dramatic reduction of computational overhead by employing coarse grids while still maintaining high accuracy. In the establis...

show abstract

Section: Filtered Subgrid Model Developmentsupporting

confidence: 55%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Machine learning to assist filtered two‐fluid model development for dense gas–particle flows

2020

View full text Add to dashboard Cite

show abstract

“…A closure model for the drift velocity can be found from rewriting its definition as follows [32,39]:…”

Section: Functional Closures (Sa-tfm)mentioning

confidence: 99%

Verification and Validation of Spatially Averaged Models for Fluidized Gas‐Particle Suspensions

Schneiderbauer

2020

Chem Eng & Technol

View full text Add to dashboard Cite

Two different methods for closure modeling of the unresolved terms appearing in the filtered two-fluid model are discussed and compared. The spatially averaged two-fluid model is based on generalizing the concepts of large eddy simulation to gas-particle flows. In the approximate deconvolution method-two-fluid model approach, the unresolved terms are modeled by an approximate deconvolution method, where an approximation of the unfiltered solution is obtained by repeated filtering. Finally, these models are applied to a lab-scale and a pilot-scale fluidized bed. Both approaches yield fairly good agreement with a highly resolved reference simulation as well as with experimental data. Additionally, both methods deliver reasonable grid-independent solutions up to a grid resolution of 2 cm in the case of Geldart type A particles.

show abstract

A gas pressure gradient‐dependent subgrid drift velocity model for drag prediction in fluidized gas–particle flows

2019

View full text Add to dashboard Cite

Due to the linear correlation between the subgrid drift velocity and the filtered drag force, modeling the drift velocity would be an alternative way to obtain the filtered drag force for coarse‐grid simulations. This work aims to improve the predictability of models for the drift velocity using a new effective marker, the filtered gas pressure gradient, which is identified by momentum balance analysis. New models are constructed based on conditional averaging of the results obtained from fine‐grid two‐fluid model simulations of three‐dimensional unbounded fluidized systems. A priori assessment is presented with the comparison between the proposed models and the best available Smagorinsky‐type model with dynamic adjustment technique proposed in the literature. Results show that the proposed models give satisfactory performance. More important, the proposed models are demonstrated to have a better adaptability for cases under various physical conditions than the Smagorinsky‐type model.

show abstract

On the choice of closure complexity in anisotropic drag closures for filtered Two Fluid Models

Cited by 41 publications

References 36 publications

Machine learning to assist filtered two‐fluid model development for dense gas–particle flows

Machine learning to assist filtered two‐fluid model development for dense gas–particle flows

Verification and Validation of Spatially Averaged Models for Fluidized Gas‐Particle Suspensions

A gas pressure gradient‐dependent subgrid drift velocity model for drag prediction in fluidized gas–particle flows

Contact Info

Product

Resources

About