A supervised machine learning approach for predicting variable drag forces on spherical particles in suspension

He, Long; Tafti, Danesh K.

doi:10.1016/j.powtec.2019.01.013

Cited by 68 publications

(29 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is widely known and accepted in physics that the drag force on each particle in fluid-particle systems, such as the one being considered in this article, is influenced strongly by the pressure and velocity fields acting on the particles. 2 Hence, we wish to explicitly model the pressure and velocity fields around a particle, in addition to the main problem of predicting its drag force. To this end, we design two multitask models, DNN-MT-Pres and DNN-MT-Vel, as described in the Experimental Setup section.…”

Section: Physics-guided Auxiliary Task Selectionmentioning

confidence: 99%

“…It is well accepted in theory that for high Reynolds numbers, the proportion of the shear components of drag (F S ) decreases. 2 To evaluate this, we consider the ratio of the magnitude of the predicted pressure components in the x direction (F P…”

Section: Physics-guided Network For Drag Force Prediction 441mentioning

confidence: 99%

“…In this article, we present PhyNet and attempt to bridge the gap between physics-based models and ML models by incorporating domain knowledge in the design and learning of ML models. Specifically, we present three ways for incorporating domain knowledge in neural networks: (1) physics-guided design of neural network architectures, (2) learning with auxiliary tasks involving physical intermediate variables, and (3) physics-guided aggregate supervision of neural network training. Our PhyNet model leverages prior physics theory to learn better representations of the drag forces affecting different particles in assemblies.…”

Section: Introductionmentioning

confidence: 99%

“…This is described in the Proposed PhyNet Framework section and in the Data set Description section. (2) We perform novel experiments to demonstrate the ability of PhyNet to interpolate (see results in the Physics-Guided Auxiliary Task Selection section, the Physics-Guided Learning Architecture section, the Performance with Limited Data section, and the Characterizing PhyNet Performance for Different (Re, /) Settings section) and to extrapolate to unseen particle assemblies (see results in the Effect of Neighborhood Size and Extrapolation to Unseen Assemblies section) and compare model performance of PhyNet to state-of-the-art baselines. (3) We characterize the model performance of Phy-Net with increase in granularity of sampled pressure and velocity fields and also the effect of change in particle neighborhood size on the model performance of PhyNet, as described in the Effect of Pressure and Velocity Sampling Methodology section and in the Effect of Neighborhood Size and Extrapolation to Unseen Assemblies section.…”

Section: Introductionmentioning

confidence: 99%

“…The Proposed PhyNet Framework section presents our problem formulation and our proposed PhyNet model. The Data set Description section describes details of the approach used for data generation (further details about the data generation process are provided in He et al, 1 He et al 2 ). The Experimental Results section presents our experimental results while the Conclusion section presents concluding remarks.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Physics-Guided Deep Learning for Drag Force Prediction in Dense Fluid-Particulate Systems

Muralidhar

Cao

et al. 2020

Big Data

Self Cite

View full text Add to dashboard Cite

Physics-based simulations are often used to model and understand complex physical systems in domains such as fluid dynamics. Such simulations, although used frequently, often suffer from inaccurate or incomplete representations either due to their high computational costs or due to lack of complete physical knowledge of the system. In such situations, it is useful to employ machine learning (ML) to fill the gap by learning a model of the complex physical process directly from simulation data. However, as data generation through simulations is costly, we need to develop models being cognizant of data paucity issues. In such scenarios, it is helpful if the rich physical knowledge of the application domain is incorporated in the architectural design of ML models. We can also use information from physics-based simulations to guide the learning process using aggregate supervision to favorably constrain the learning process. In this article, we propose PhyNet, a deep learning model using physics-guided structural priors and physics-guided aggregate supervision for modeling the drag forces acting on each particle in a computational fluid dynamics-discrete element method. We conduct extensive experiments in the context of drag force prediction and showcase the usefulness of including physics knowledge in our deep learning formulation. PhyNet has been compared with several state-of-the-art models and achieves a significant performance improvement of 7.09% on average. The source code has been made available*.

show abstract

Section: Physics-guided Auxiliary Task Selectionmentioning

confidence: 99%

Section: Physics-guided Network For Drag Force Prediction 441mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

mentioning

confidence: 99%

See 3 more Smart Citations

Physics-Guided Deep Learning for Drag Force Prediction in Dense Fluid-Particulate Systems

Muralidhar

Cao

et al. 2020

Big Data

Self Cite

View full text Add to dashboard Cite

show abstract

A study of methanol‐to‐olefins packed bed reactor performance using particle‐resolved CFD and machine learning

Zhu,

Kenig

2024

AIChE Journal

View full text Add to dashboard Cite

In this study, particle‐resolved computational fluid dynamics (CFD) simulations were performed to analyze fluid flow, mass transport, and reaction phenomena in methanol‐to‐olefins packed bed reactors with diverse cylindrical configurations and operating conditions. Utilizing validated CFD data, data‐driven surrogate models were developed based on several representative machine learning (ML) techniques. Comprehensive training and optimization of ML model hyperparameters were performed, followed by a comparative assessment of their capabilities to predict reactor performance. Subsequently, data‐driven surrogate models together with CFD simulations were applied to optimize catalyst structure design and operating conditions. Finally, a hybrid approach was developed that couples the ML‐aided data‐driven model with a genetic algorithm‐based multi‐objective optimization. The resulting hybrid method was applied to find the Pareto‐optimal compromise between pressure drop and light olefins yield.

show abstract

Toward particle-resolved accuracy in Euler–Lagrange simulations of multiphase flow using machine learning and pairwise interaction extended point-particle (PIEP) approximation

Balachandar

Moore

Akiki

et al. 2020

Theor. Comput. Fluid Dyn.

View full text Add to dashboard Cite

A supervised machine learning approach for predicting variable drag forces on spherical particles in suspension

Cited by 68 publications

References 24 publications

Physics-Guided Deep Learning for Drag Force Prediction in Dense Fluid-Particulate Systems

Physics-Guided Deep Learning for Drag Force Prediction in Dense Fluid-Particulate Systems

A study of methanol‐to‐olefins packed bed reactor performance using particle‐resolved CFD and machine learning

Toward particle-resolved accuracy in Euler–Lagrange simulations of multiphase flow using machine learning and pairwise interaction extended point-particle (PIEP) approximation

Contact Info

Product

Resources

About