Data-driven modeling of coarse mesh turbulence for reactor transient analysis using convolutional recurrent neural networks

Liu, Yang; Hu, Rui; Kraus, Adam; Balaprakash, Prasanna; Obabko, Aleksandr

doi:10.1016/j.nucengdes.2022.111716

Cited by 12 publications

(6 citation statements)

References 56 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Altogether, compared with highly resolved simulations, coarse simulations are much faster and satisfy the short turnover duration required in engineering decisions. 128 Despite its benefit to remarkably reduce the computational cost, in fact coarse simulations still need huge computational resources for largescale flow devices. For example, in industrial gas−solid FCC riser flow reactors, the mesh resolution requirements are ∼10 −3 m, which is computationally prohibitive for reactor-scale simulations (10°∼ 10 1 m).…”

Section: Modeling Of Turbulence Stress Model Coefficientsmentioning

confidence: 99%

“…The reported method can accelerate DEM computation speed by orders of magnitude. Altogether, compared with highly resolved simulations, coarse simulations are much faster and satisfy the short turnover duration required in engineering decisions . Despite its benefit to remarkably reduce the computational cost, in fact coarse simulations still need huge computational resources for large-scale flow devices.…”

Section: Current Status and Challengesmentioning

confidence: 99%

See 1 more Smart Citation

Review of Machine Learning for Hydrodynamics, Transport, and Reactions in Multiphase Flows and Reactors

Zhu

Chen

Ouyang

et al. 2022

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

Artificial intelligence (AI), machine learning (ML), and data science are leading to a promising transformative paradigm. ML, especially deep learning and physics-informed ML, is a valuable toolkit that complements incomplete domain-specific knowledge in conventional experimental and computational methods. ML can provide flexible techniques to facilitate the conceptual development of new robust predictive models for multiphase flows and reactors by finding hidden pattern/information/mechanism in a data set. Due to such emergence, we thereby comprehensively survey, explore, analyze, and discuss key advancements of recent ML applications to hydrodynamics, heat and mass transfer, and reactions in single-phase and multiphase flow systems from different aspects: (1) development of multiphase closure models of drag force, turbulence stresses and heat/mass transfer to improve the accuracy and efficiency of typical CFD simulations; (2) image reconstruction, regime identification, key parameter predictions, and optimization of multiphase flow and transport fields; (3) reaction kinetics modeling (e.g., predictions of reaction networks, kinetic parameters, and species production) and reaction condition optimization. These sections also discuss and analyze the key advantages and weakness of ML for solving the problems in the domain of multiphase flows and reactors. Finally, we summarize the under-solving challenges and opportunities in order to identify future directions that would be useful for the research community. Future development and study of multiphase flows and reactors are envisaged to be accelerated by ML and data science.

show abstract

Section: Modeling Of Turbulence Stress Model Coefficientsmentioning

confidence: 99%

Section: Current Status and Challengesmentioning

confidence: 99%

Review of Machine Learning for Hydrodynamics, Transport, and Reactions in Multiphase Flows and Reactors

Zhu

Chen

Ouyang

et al. 2022

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

show abstract

“…Yang et al . studied the capability of various artificial intelligence algorithms for reactor transient analysis based on computational fluid dynamics (CFD) data 17 – 19 . Moreover, Wang et al .…”

Section: Background and Summarymentioning

confidence: 99%

An open time-series simulated dataset covering various accidents for nuclear power plants

Xiao

Liang

et al. 2022

Sci Data

View full text Add to dashboard Cite

Nuclear energy plays an important role in global energy supply, especially as a key low-carbon source of power. However, safe operation is very critical in nuclear power plants (NPPs). Given the significant impact of human-caused errors on three serious nuclear accidents in history, artificial intelligence (AI) has increasingly been used in assisting operators with regard to making various decisions. In particular, data-driven AI algorithms have been used to identify the presence of accidents and their root causes. However, there is a lack of an open NPP accident dataset for measuring the performance of various algorithms, which is very challenging. This paper presents a first-of-its-kind open dataset created using PCTRAN, a pre-developed and widely used simulator for NPPs. The dataset, namely nuclear power plant accident data (NPPAD), basically covers the common types of accidents in typical pressurised water reactor NPPs, and it contains time-series data on the status or actions of various subsystems, accident types, and severity information. Moreover, the dataset incorporates other simulation data (e.g., radionuclide data) for conducting research beyond accident diagnosis.

show abstract

“…We refer these two search methods as RD+ACBO and GP+ACBO because they both follow a centralized asynchronous architecture. DeepHyper has been used to improve the accuracy of neural networks in several scientific machine learning applications [38], [39], [40], [41]. In addition, we implemented an asynchronous variant of Hyperband (ASHA) [22], [42] based on the pruner API of the Optuna Python package.…”

Section: E Application To Neural Network Hyperparameter Tuningmentioning

confidence: 99%

Asynchronous Distributed Bayesian Optimization at HPC Scale

Égelé¹,

Gouneau²,

Vishwanath³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Bayesian optimization (BO) is a widely used approach for computationally expensive black-box optimization such as simulator calibration and hyperparameter optimization of deep learning methods. In BO, a dynamically updated computationally cheap surrogate model is employed to learn the input-output relationship of the black-box function; this surrogate model is used to explore and exploit the promising regions of the input space. Multipoint BO methods adopt a single manager/multiple workers strategy to achieve high-quality solutions in shorter time. However, the computational overhead in multipoint generation schemes is a major bottleneck in designing BO methods that can scale to thousands of workers. We present an asynchronous-distributed BO (ADBO) method wherein each worker runs a search and asynchronously communicates the input-output values of black-box evaluations from all other workers without the manager. We scale our method up to 4,096 workers and demonstrate improvement in the quality of the solution and faster convergence. We demonstrate the effectiveness of our approach for tuning the hyperparameters of neural networks from the Exascale computing project CANDLE benchmarks.

show abstract

Data-driven modeling of coarse mesh turbulence for reactor transient analysis using convolutional recurrent neural networks

Cited by 12 publications

References 56 publications

Review of Machine Learning for Hydrodynamics, Transport, and Reactions in Multiphase Flows and Reactors

Review of Machine Learning for Hydrodynamics, Transport, and Reactions in Multiphase Flows and Reactors

An open time-series simulated dataset covering various accidents for nuclear power plants

Asynchronous Distributed Bayesian Optimization at HPC Scale

Contact Info

Product

Resources

About