Deep learning of turbulent scalar mixing

Raissi, Maziar; Babaee, Hessam; Givi, Peyman

doi:10.1103/physrevfluids.4.124501

Cited by 66 publications

(32 citation statements)

References 85 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As far as the presumed PDF models for LES are concerned, the DNS evidence is used to develop training data and different algorithms are evaluated in [102]. Another paper, directly related to turbulent mixing, is [103]. Using deep learning ideas, the authors propose a framework to deduce models from the experimental measurements of the PDF.…”

Section: Discussionmentioning

confidence: 99%

Mixing in Turbulent Flows: An Overview of Physics and Modelling

Pozorski

Wacławczyk

2020

Processes

View full text Add to dashboard Cite

Turbulent flows featuring additional scalar fields, such as chemical species or temperature, are common in environmental and industrial applications. Their physics is complex because of a broad range of scales involved; hence, efficient computational approaches remain a challenge. In this paper, we present an overview of such flows (with no particular emphasis on combustion, however) and we recall the major types of micro-mixing models developed within the statistical approaches to turbulence (the probability density function approach) as well as in the large-eddy simulation context (the filtered density function). We also report on some trends in algorithm development with respect to the recent progress in computing technology.

show abstract

Section: Discussionmentioning

confidence: 99%

Mixing in Turbulent Flows: An Overview of Physics and Modelling

Pozorski

Wacławczyk

2020

Processes

View full text Add to dashboard Cite

show abstract

“…PINNs were first introduced in 16 and since then PINNs have been successfully applied in solving forward and inverse problems in many practical applications 41 –48. The analysis and convergence study of PINNs have been carried out in 49 .…”

Section: Methodsmentioning

confidence: 99%

Identifiability and predictability of integer- and fractional-order epidemiological models using physics-informed neural networks

Kharazmi

Cai

Zheng

et al. 2021

Preprint

View full text Add to dashboard Cite

We analyze a plurality of epidemiological models through the lens of physics-informed neural networks (PINNs) that enable us to identify multiple time-dependent parameters and to discover new data-driven fractional differential operators. In particular, we consider several variations of the classical susceptible-infectious-removed (SIR) model by introducing more compartments and delay in the dynamics described by integer-order, fractional-order, and time-delay models. We report the results for the spread of COVID-19 in New York City, Rhode Island and Michigan states, and Italy, by simultaneously inferring the unknown parameters and the unobserved dynamics. For integer-order and time-delay models, we fit the available data by identifying time-dependent parameters, which are represented by neural networks (NNs). In contrast, for fractional differential models, we fit the data by determining different time-dependent derivative orders for each compartment, which we represent by NNs. We investigate the identifiability of these unknown functions for different datasets, and quantify the uncertainty associated with NNs and with control measures in forecasting the pandemic.

show abstract

“…In this example, detailed in Ref. [37], we consider the problem of mixing of a Fickian passive scalar = ( , ) ( denotes time and is the position vector), with diffusion coefficient Γ from an initially symmetric binary state within the bounds −1 ≤ ≤ +1. Therefore, the single-point PDF of at the initial time is (0, ) = Modern ML techniques have the potential to be utilized for PDF model developments [39,40].…”

Section: A Deep-learning Of Turbulent Scalar Mixingmentioning

confidence: 99%

“…Figure 1-Physics-Informed Neural Network: The residual neural network f is obtained by approximating the unknown solution u by a deep neural network and by taking the required spatial and temporal derivatives using automatic differentiation.Taken from Ref [37]. …”

mentioning

confidence: 99%

Computational Turbulent Combustion in the Age of Artificial Intelligence and Quantum Information

Givi¹

2021

ijmph

Self Cite

View full text Add to dashboard Cite

The impact of high-performance computing on the society has been enormous, but it is easy to be taken for granted. In today's world, it is virtually impossible to imagine system design or major decision making not aided via predictive modeling and simulation. Now that we are experiencing the Data Revolution and the emergence of the Second Quantum Revolution, it is wise to consider both of these elements in computational science and engineering. Data-driven modeling approaches and demonstrated speed-ups of quantum algorithms have the potential to transform scientific discovery. This will affect the fabrics of industrialized societies in diverse disciplines. A research arena which can substantially benefit from these technologies is combustion. This field has been the subject of heavy computational research for many decades now. In this review, some examples taken from the previous works of the author are presented to demonstrate how the field of computational turbulent combustion is benefiting from modern developments in machine learning (ML) and quantum computing (QC).

show abstract

Deep learning of turbulent scalar mixing

Cited by 66 publications

References 85 publications

Mixing in Turbulent Flows: An Overview of Physics and Modelling

Mixing in Turbulent Flows: An Overview of Physics and Modelling

Identifiability and predictability of integer- and fractional-order epidemiological models using physics-informed neural networks

Computational Turbulent Combustion in the Age of Artificial Intelligence and Quantum Information

Contact Info

Product

Resources

About