FlowDNN: a physics-informed deep neural network for fast and accurate flow prediction

Chen, Donglin; Gao, Xiang; Xu, Chuanfu; Wang, Siqi; Chen, Shizhao; Fang, Jianbin; Wang, Zheng

doi:10.1631/fitee.2000435

Cited by 26 publications

(8 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Current developments show that classical CFD simulations could be complemented or replaced by other techniques. Machine learning techniques can be used to accelerate simulations or to improve turbulence modelling [129], and physics-informed neural networks (PINN) are increasingly used, for their ability to perform calculations 200 times faster to the same degree of accuracy [130,131]. Another technique, which currently has no real application for bioreactor modelling, is quantum CFD (QCFD) [132][133][134][135].…”

Section: Hardware and Softwarementioning

confidence: 99%

Computational Fluid Dynamics for Advanced Characterisation of Bioreactors Used in the Biopharmaceutical Industry – Part I: Literature Review

Seidel¹,

Schirmer²,

Maschke³

et al. 2023

Computational Fluid Dynamics - Recent Advances, New Perspectives and Applications

View full text Add to dashboard Cite

Computational fluid dynamics (CFD) is a widely used tool for investigating fluid flows in bioreactors. It has been used in the biopharmaceutical industry for years and has established itself as an important tool for process engineering characterisation. As a result, CFD simulations are increasingly being used to complement classical process engineering investigations in the laboratory with spatially and temporally resolved results, or even replace them when laboratory investigations are not possible. Parameters that can be determined include the specific power input, Kolmogorov length, hydrodynamic stress, mixing time, oxygen transfer rate, and for cultivations with microcarriers, the NS1 criterion. In the first part of this series, a literature review illustrates how these parameters can be determined using CFD and how they can be validated experimentally. In addition, an overview of the hardware and software typically used for bioreactor characterisation will also be provided, including process engineering parameter investigations from the literature. In the second part of this series, the authors’ research results will be used to show how the process engineering characterisation of mechanically driven bioreactors for the biopharmaceutical industry (stirred, orbitally shaken, and wave-mixed) can be determined and validated using CFD.

show abstract

Section: Hardware and Softwarementioning

confidence: 99%

Computational Fluid Dynamics for Advanced Characterisation of Bioreactors Used in the Biopharmaceutical Industry – Part I: Literature Review

Seidel¹,

Schirmer²,

Maschke³

et al. 2023

Computational Fluid Dynamics - Recent Advances, New Perspectives and Applications

View full text Add to dashboard Cite

show abstract

“…Moreover, the loss function can be used to inform the data-driven models of existing physical laws to be conserved, such as conservation of mass and momentum, so their predictions are physically realistic and accurate. Examples for this method are the physics-informed deep neural network (FlowDNN) by Chen et al (2022) and the physicsinformed neural network (PINN) by Raissi et al (2019).…”

Section: Data-driven Modelsmentioning

confidence: 99%

Recent advances in modeling turbulent wind flow at pedestrian-level in the built environment

Zhong

Liu

Zhao

et al. 2022

ARIN

View full text Add to dashboard Cite

Pressing problems in urban ventilation and thermal comfort affecting pedestrians related to current urban development and densification are increasingly dealt with from the perspective of climate change adaptation strategies. In recent research efforts, the prime objective is to accurately assess pedestrian-level wind (PLW) environments by using different simulation approaches that have reasonable computational time. This review aims to provide insights into the most recent PLW studies that use both established and data-driven simulation approaches during the last 5 years, covering 215 articles using computational fluid dynamics (CFD) and typical data-driven models. We observe that steady-state Reynolds-averaged Navier-Stokes (SRANS) simulations are still the most dominantly used approach. Due to the model uncertainty embedded in the SRANS approach, a sensitivity test is recommended as a remedial measure for using SRANS. Another noted thriving trend is conducting unsteady-state simulations using high-efficiency methods. Specifically, both the massively parallelized large-eddy simulation (LES) and hybrid LES-RANS offer high computational efficiency and accuracy. While data-driven models are in general believed to be more computationally efficient in predicting PLW dynamics, they in fact still call for substantial computational resources and efforts if the time for development, training and validation of a data-driven model is taken into account. The synthesized understanding of these modeling approaches is expected to facilitate the choosing of proper simulation approaches for PLW environment studies, to ultimately serving urban planning and building designs with respect to pedestrian comfort and urban ventilation assessment.

show abstract

“…PINNs are neural networks that learn distribution patterns of training data to approximate the physical laws. It has been applied in a range of fields, such as materials [36,37], mechanics [38,39], fluids [40,41,42] and bioengineering [43,44]. PINNs incorporate the physical laws governing scientific data in the deep learning framework, resulting in improved accuracy and credibility.…”

Section: Volumetric Super-resolution Of Scientific Datamentioning

confidence: 99%

Estimation and Prediction of Typhoons and Wave Overtopping in Qingdao, China

Wang

Hao

et al. 2020

J. Ocean Univ. China

View full text Add to dashboard Cite

When introducing physics-constrained deep learning solutions to the volumetric super-resolution of scientific data, the training is challenging to converge and always time-consuming. We propose a new hierarchical sampling method based on octree to solve these difficulties. In our approach, scientific data is preprocessed before training, and a hierarchical octreebased data structure is built to guide sampling on the latent context grid. Each leaf node in the octree corresponds to an indivisible subblock of the volumetric data. The dimensions of the subblocks are different, making the number of sample points in each randomly cropped training data block to be adaptive. We reconstruct the octree at intervals according to loss distribution to perform the multi-stage training. With the Rayleigh-Bénard convection problem, we deploy our method to state-of-the-art models. We constructed adequate experiments to evaluate the training performance and model accuracy of our method. Experiments indicate that our sampling optimization improves the convergence performance of physics-constrained deep learning super-resolution solutions. Furthermore, the sample points and training time are significantly reduced with no drop in model accuracy. We also test our method in training tasks of other deep neural networks, and the results show our sampling optimization has extensive effectiveness and applicability. The code is publicly available at https://github.com/xinjiewang/octree-based sampling.

show abstract

FlowDNN: a physics-informed deep neural network for fast and accurate flow prediction

Cited by 26 publications

References 27 publications

Computational Fluid Dynamics for Advanced Characterisation of Bioreactors Used in the Biopharmaceutical Industry – Part I: Literature Review

Computational Fluid Dynamics for Advanced Characterisation of Bioreactors Used in the Biopharmaceutical Industry – Part I: Literature Review

Recent advances in modeling turbulent wind flow at pedestrian-level in the built environment

Estimation and Prediction of Typhoons and Wave Overtopping in Qingdao, China

Contact Info

Product

Resources

About