Metallurgical reactors, especially in ironmaking/steelmaking process, characterise with high-temperature turbulence, multiphase flow, mass/heat transfer and reactions. Computational fluid dynamics (CFD) simulation-based design and optimisation are of significance for efficient metallurgical performance. However, the difficulty and cost to numerically solve the nonlinear controlling equations combined with data pre/post-processing make the whole CFD simulation process time-consuming, which makes it challenging to provide in-time feedback for industrial practices. The popularisation and prosperous development of machine learning bring new opportunities for promoting CFD performance. Discussion has been made on the current research progress of applying machine learning in the whole CFD workflow including pre-processing, solving, and post-processing. Among them, the time consumed by manual pre-processing exceeds 50% of CFD tasks in general. The machine learning or parametric modelling methods can reduce pre-processing time by three orders in the estimate. The solving step is expected to be accelerated by 5 to 1000 times using machine learning. A brief review of machine learning coupled CFD is provided, as is a prospective on its development. Discussion is presented on the main functions, challenges, typical techniques and future directions of applying machine learning in CFD simulation of metallurgical reactors, for the purpose of making CFD faster, more accurate, and better visualised based on the metallurgical practices.
ZhuLTChenXZOuyangB, et al.Review of machine learning for hydrodynamics, transport, and reactions in multiphase flows and reactors. Ind Eng Chem Res2022; 61: 9901–9949.
11.
SunJKZhangJSJiangR, et al.Numerical simulation of droplet splashing behavior in steelmaking converter based on VOF-to-DPM hybrid model and AMR technique. Metall Mater Trans B2024; 55: 1098–1116.
12.
HeBCXinZCZhangJS, et al.A computational fluid dynamics – thermodynamics coupled approach to simulate desulfurization in ladle furnace based on interface equilibrium assumption. Steel Res Int2023: 94: 2200965.
13.
ZhangJSQinBMLiuYH, et al.Multiphase flow inside a four-strand continuous casting tundish using three types of ladle shrouds. J Iron Steel Res Int2023; 30: 1171–1181.
14.
CuiHNZhangJSXinZC, et al.Large eddy simulation of the effect of EMBr on the flow behavior of thin slab continuous casting mold with ultra-high casting speed. Continuous Casting2023; 3: 33–41.
15.
ChenJQ. Computational fluid dynamic 2035 study vision in China. 1st ed. Beijing. China: Science Press, China Science Publishing and Media Ltd, 2023, p.56.
16.
PietroniNCampenMShefferA, et al.Hex-mesh generation and processing: A survey. ACM Trans Graph2023; 42: 1–44.
17.
BlackeT. Meeting the challenge for automated conformal hexahedral meshing. In: 9th international meshing roundtable, IMR 2000, New Orleans, Louisiana, USA, October 2–5 October 2000, pp.11–20.
18.
YaoSYanBChenB, et al.An ANN-based element extraction method for automatic mesh generation. Expert Syst Appl2005; 29: 193–206.
19.
LuPWangNHChangXH, et al.An automatic isotropic/anisotropic hybrid grid generation technique for viscous flow simulations based on an artificial neural network. Chin J Aeronaut2022; 35: 102–117.
20.
HuangKKrügenerMBrownA, et al.Machine learning-based optimal mesh generation in computational fluid dynamics. arxiv Preprint arxiv:2102.12923, 2021.
21.
WuTFLiuXJAnW, et al.A mesh optimization method using machine learning technique and variational mesh adaptation. Chin J Aeronaut2022; 35: 27–41.
22.
LorsungCBaratiFA. Mesh deep Q network: A deep reinforcement learning framework for improving meshes in computational fluid dynamics. AIP Adv2023; 13: 015026.
23.
OjhaVChenGFidkowskiK. Initial mesh generation for solution-adaptive methods using machine learning. In: AIAA SCITECH 2022 Forum, 7 January 2022, p. 1244. San Diego, CA and Online: AIAA, USA.
24.
ChenXHLiuJZhangQY, et al.Developing a novel structured mesh generation method based on deep neural networks. Phys Fluids2023; 35: 097137.
25.
JumperJEvansRPritzelA, et al.Highly accurate protein structure prediction with AlphaFold. Nature2021; 596: 583–589.
26.
VivienM. Method of the year: Protein structure prediction. Nat Methods2022; 19: 5–10.
27.
ZhangSTongHHXuJJ, et al.Graph convolutional networks: A comprehensive review. Comput Social Netw2019; 6: 1–23.
28.
JinBZXuXJ. Contemporaneous causality among price indices of ten major steel products. Ironmak Steelmak2024; 51: 515–526.
29.
JinBZXuXJ. Gaussian process regression based silver price forecasts. J Uncertain Syst2024. DOI: 10.1142/S1752890924500132
30.
JinBZXuXJ. Price forecasting through neural networks for crude oil, heating oil, and natural gas. Measurement: Energy2024; 1: 100001.
31.
ChoiHMoinP. Grid-point requirements for large eddy simulation: Chapman’s estimates revisited. Phys Fluids2012; 24: 011702.
32.
StevensBColoniusT. FiniteNet: A fully convolutional LSTM network architecture for time-dependent partial differential equations. Arxiv Preprint Arxiv2002.03014 2020.
33.
WeatherittJSandbergR. A novel evolutionary algorithm applied to algebraic modifications of the RANS stress–strain relationship. J Comput Phys2016; 325: 22–37.
RaissiMPerdikarisPKarniadakisGE. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J Comput Phys2019; 378: 686–707.
36.
RenXHuPSuH, et al.Physics-informed neural networks for transonic flow around a cylinder with high Reynolds number. Phys Fluids2024; 36: 036129.
37.
ZhaoZStuebnerMLuaJ, et al.Full-field temperature recovery during water quenching processes via physics-informed machine learning. J Mater Process Technol2022; 303: 117534.
38.
SharmaRGuoBYRaissiM, et al.Physics-informed machine learning of argon gas-driven melt pool dynamics. J Manuf Sci Eng2024; 146: 081008–1.
TairaKHematiMSBrutonSL, et al.Modal analysis of fluid flows: Applications and outlook. AIAA J2020; 58: 998–1022.
41.
Abadía-HerediaRLópez-MartínMCarroB, et al.A predictive hybrid reduced order model based on proper orthogonal decomposition combined with deep learning architectures. Expert Syst Appl2022; 187: 115910.
42.
TakeishiNKawaharaYYairiT. Learning Koopman invariant subspaces for dynamic mode decomposition. Adv Neural Inf Process Syst2017; 30: 1130–1140.
43.
EivaziHLe ClaincheSHoyasS, et al.Towards extraction of orthogonal and parsimonious non-linear modes from turbulent flows. Expert Syst Appl2022; 202: 117038.
44.
MurataTFukamiKFukagataK. Nonlinear mode decomposition with convolutional neural networks for fluid dynamics. J Fluid Mech2020; 882: A13.
45.
BrendanDTDuraisamyKAlonsoJJ. A machine learning strategy to assist turbulence model development. In: 53rd AIAA aerospace sciences meeting, 5 January–9 January 2015, p.1287. Kissimmee, Florida: AIAA, USA.
46.
BeckAFladDMunzCD. Deep neural networks for data-driven LES closure models. J Comput Phys2019; 398: 108910.
47.
ThuereyNWeißenowKPrantlL, et al.Deep learning methods for Reynolds-averaged Navier–Stokes simulations of airfoil flows. AIAA J2020; 58: 25–36.
48.
CaiSZMaoZPWangZC, et al.Physics-informed neural networks (PINNs) for fluid mechanics: A review. Acta Mech Sin2021; 37: 1727–1738.
49.
BikakisNChrysanthisPKPapastefanatosG, et al.Special issue on machine learning approaches in big data visualization. IEEE Comput Graph Appl2022; 42: 39–40.
50.
LiuCJiangRWeiD, et al.Deep learning approaches in flow visualization. Adv Aerodyn2022; 4: 1–14.
51.
GaoHSunLWangJX. Super-resolution and denoising of fluid flow using physics-informed convolutional neural networks without high-resolution labels. Phys Fluids2021; 33: 073603.
52.
HongFLiuCYuanX. DNN-VolVis: Interactive volume visualization supported by deep neural network. In: 2019 IEEE Pacific visualization symposium (PacificVis): IEEE, 23 April–26 April 26 2019, pp. 282–291. Bangkok, Thailand: IEEE New York City, U.S.
53.
GuoLYeSHanJ, et al.SSR-VFD: Spatial super-resolution for vector field data analysis and visualization. In: Proceedings of IEEE Pacific visualization symposium, 3 June–5 June 2020. Tianjin, China: IEEE New York City, U.S.
54.
HöhleinKKernMHewsonT, et al.A comparative study of convolutional neural network models for wind field downscaling. Meteorol Appl2020; 27: e1961.
55.
FranzKRoscherRMiliotoA, et al.Ocean eddy identification and tracking using neural networks. In: IGARSS 2018 IEEE international geoscience and remote sensing symposium: IEEE, 22 July 2018, pp.6887–6890.
56.
LiuYLuYWangY, et al.A CNN-based shock detection method in flow visualization. Comput Fluids2019; 184: 1–9.
57.
HanJTaoJWangC. Flownet: A deep learning framework for clustering and selection of streamlines and stream surfaces. IEEE Trans Visual Comput Graph2018; 26: 1732–1744.
58.
ZhangSYLiHWangK, et al.Accelerating intelligent microfluidic image processing with transfer deep learning: A microchannel droplet/bubble breakup case study. Sep Purif Technol2023; 315: 123703.
59.
GuoYLiuYOerlemansA, et al.Deep learning for visual understanding: A review. Neurocomputing2016; 187: 27–48.
BenderEMGebruTMcMillan-MajorA, et al.On the dangers of stochastic parrots: Can language models be too big? In: Proceedings of the 2021 ACM conference on fairness, Virtual Event Canada, 3 March–10 March 2021, pp.610–623. New York, NY, USA: Association for Computing Machinery.
