Point-particle drag, lift, and torque closure models using machine learning: Hierarchical approach and interpretability

Siddani, Bhargav Sriram; Balachandar, S.

doi:10.1103/physrevfluids.8.014303

Cited by 22 publications

(17 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is worth noting that the test cases in this study only extend up to Φ = 0.4 and the effectiveness of the model at higher Φ has not been explored. We also note that the existing literature 34,36,39 on this topic does not examine any case beyond Φ = 0.4 either. Nevertheless, it may remain essential to investigate the effectiveness of the model at Φ values exceeding 0.4, considering the potential formation and disintegration of particle clusters driven by relative motion between particles.…”

Section: Interpolation Of Mpp Modelmentioning

confidence: 93%

“…This suboptimal performance could be addressed in future investigations. Recently, machine learning methods have shown promises in addressing complex challenges in CFD, 39,40 and they could serve as an alternative approach for estimating the 2M coefficients. In addition, in E-L simulations, a mean correlation derived from flow through arrays of stationary particles is commonly used to approximate the force on moving particles, especially in gas-solid flows with high density ratios.…”

Section: Conclusion and Remarksmentioning

confidence: 99%

See 1 more Smart Citation

Interpolation of probability‐driven model to predict hydrodynamic forces and torques in particle‐laden flows

Zhu

Wachs

2023

AIChE Journal

View full text Add to dashboard Cite

The development of hydrodynamic force/torque closure models with physical fidelity is crucial for ensuring reliable Euler–Lagrange simulations in particle‐laden flows. Our previous work (Seyed‐Ahmadi and Wachs. J Fluid Mech. 2020;900:A21) proposed a microstructure‐informed probability‐driven point‐particle (MPP) method to construct a data‐driven particle‐position‐dependent closure model, incorporating the effect of surrounding particle positions on forces/torques. However, the MPP model is not pluggable in Euler–Lagrange simulations due to the computation of constant coefficients through linear regression and reliance on statistical arguments to obtain the probability map for a pair of values of solid volume fraction (Φ) and Reynolds number (Re). To overcome this limitation, we propose an interpolated MPP (iMPP) method, involving interpolation in the Φ and Re spaces. Our results demonstrate that the iMPP method can capture over 70% of the total fluctuations in hydrodynamic forces/torques in approximately 97.8% of the tested cases. This advancement contributes to a more versatile closure model suitable for integration into E‐L simulations.

show abstract

Section: Interpolation Of Mpp Modelmentioning

confidence: 93%

Section: Conclusion and Remarksmentioning

confidence: 99%

Interpolation of probability‐driven model to predict hydrodynamic forces and torques in particle‐laden flows

Zhu

Wachs

2023

AIChE Journal

View full text Add to dashboard Cite

show abstract

“…Furthermore, its magnitude is substantially smaller than the transverse torque components and therefore harder to predict. The results presented in 24 indicate that the generalizability for this component is low especially for high Reynolds numbers, which suggests the need for more training data.…”

Section: Uniformly Random Distributionmentioning

confidence: 96%

“…Drawing inspiration from multi-body approaches in physics and molecular dynamics 29,30 a hierarchical machine learning approach that systematically includes higher-order interactions in a step-by-step manner can be considered. In the hierarchical framework, the hydrodynamic force experienced by an i th particle is represented by the following series expansion 24 :…”

Section: Hierarchical Machine Learning Approachmentioning

confidence: 99%

“…The binary and ternary force/torque maps that we plan to use in this work (shown in Figure 1) were trained based on random uniform distributions of particles. 24 In employing the trained neural-network model to outof-training distributions, we make an important assumption that the binary and ternary force maps remain the same at the other distributions as well. Thus, any difference in the predicted force and torque distribution is due to differences in how the neighbors are spatially distributed.…”

Section: Out-of-training Generalizabilitymentioning

confidence: 99%

See 1 more Smart Citation

Investigating the influence of particle distribution on force and torque statistics using hierarchical machine learning

Siddani,

Balachandar,

Zhou

et al. 2024

AIChE Journal

Self Cite

View full text Add to dashboard Cite

An accurate representation of hydrodynamic force and torque experienced by every particle in a distribution can be obtained from particle resolved (PR) simulations. These unique quantities are influenced by the deterministic position of surrounding particles. However, systems simulated with this methodology are typically limited to particles due to the involved computational cost. This resource requirement is a major bottleneck in analyzing the effect of variations in particle distribution. This article attempts to address this bottleneck by availing relatively inexpensive deep learning models. The surrogate models that we employ in this article use a physics‐based hierarchical framework and symmetry‐preserving neural networks to achieve robustness with limited training data. This article first performs additional generalizability tests on PR data of distinct distributions that are not involved in the training process. The models are then deployed on several different particle distributions. Impact of clustering and structure on the observed statistics are investigated.

show abstract

Neural network models for preferential concentration of particles in two-dimensional turbulence

Maurel-Oujia,

Jain,

Matsuda

et al. 2024

Theor. Comput. Fluid Dyn.

View full text Add to dashboard Cite

Point-particle drag, lift, and torque closure models using machine learning: Hierarchical approach and interpretability

Cited by 22 publications

References 36 publications

Interpolation of probability‐driven model to predict hydrodynamic forces and torques in particle‐laden flows

Interpolation of probability‐driven model to predict hydrodynamic forces and torques in particle‐laden flows

Investigating the influence of particle distribution on force and torque statistics using hierarchical machine learning

Neural network models for preferential concentration of particles in two-dimensional turbulence

Contact Info

Product

Resources

About