工业互联网中数字孪生系统的机理+数据融合建模方法

doi:10.12267/j.issn.2096-5931.2022.10.008

摘要/Abstract

摘要：

工业互联网的快速发展为学术界以及工业界带来了新型研发范式——数据密集型科学发现,融合物理机理以及数据驱动的建模方法是其中的研究热点之一,这种方式可以充分发挥机理仿真可解释性和泛化能力强、数据驱动模型灵活性和可学习的优势,为未来数字孪生系统提供高效、灵活的工具和方法。通过聚焦于工业互联网中构建数字孪生系统的机理+数据融合建模方法,首先阐述了基本数学原理以及建模方法,并对比了机理+数据融合建模与传统数据模型、机理模型的区别;然后从模型选择、物理机理约束以及实际任务需求3个角度详细给出了机理+数据融合建模方法的构造过程,总结了目前学术界的最新研究进展;最后介绍了国内外关于机理+数据融合建模方法在工业设备设计优化、生产制造、运行维护方面的实际落地应用场景。

关键词: 深度学习, 机理仿真, 数据驱动, 物理启发神经网络, 数字孪生

Abstract:

The development and prosperity of the Industrial Internet have brought a novel paradigm to the academic and industrial communities - Data-intensive Scientific Discovery. The modeling method of fusion physics mechanism and data driven is one of the research hotspots,which can provide efficient and flexible analysis tools for future digital twin system. This approach can benefit from both mechanism simulation (interpretability and generalization ability) and data-driven model (flexibility and learning ability), especially in the deep learning architecture. In this context, this paper focuses on the mechanism + data fusion method in the digital twin system for Industrial Internet. Firstly, the basic mathematical principles and modeling methods are established, while the differences between mechanism + data fusion modeling and traditional models are compared. Then, machine-learning model selection, physics mechanism constraints, and actual task requirements are introduced in detail, and the recent research progress and development are summarized. Finally, the actual application scenarios of this method are reported from three perspectives, including design optimization, manufacturing, and operation maintenance.

Key words: deep learning, mechanism simulation, data driven, physics-informed neural network, digital twin

中图分类号:

李硕, 刘天源, 黄锋, 解鑫, 张金义. 工业互联网中数字孪生系统的机理+数据融合建模方法[J]. 信息通信技术与政策, 2022, 48(10): 52-61.

LI Shuo, LIU Tianyuan, HUANG Feng, XIE Xin, ZHANG Jinyi. Mechanism + data fusion modeling method in digital twin system for industrial internet[J]. Information and Communications Technology and Policy, 2022, 48(10): 52-61.

导出引用管理器 EndNote|Ris|BibTeX

链接本文:

http://ictp.caict.ac.cn/CN/10.12267/j.issn.2096-5931.2022.10.008

http://ictp.caict.ac.cn/CN/Y2022/V48/I10/52

图/表 2

参考文献 62

[1]	BRUNTON S L, NOACK B R, KOUMOUTSAKOS P. Machine learning for fluid mechanics[J]. Annual Review of Fluid Mechanics, 2020, 52(1):477-508. doi: 10.1146/annurev-fluid-010719-060214 URL
[2]	HEY T, TANSLEY S, TOLLE K. The fourth paradigm: data-intensive scientific discovery[J]. Proceedings of the IEEE, 2011, 99(8):1334-1337. doi: 10.1109/JPROC.2011.2155130 URL
[3]	LeCun Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521(7553):436-444. doi: 10.1038/nature14539 URL
[4]	JUMPER J, EVANS R, PRITZEL A, et al. Highly accurate protein structure prediction with AlphaFold[J]. Nature, 2021, 596(7873):583-589. doi: 10.1038/s41586-021-03819-2 URL
[5]	BREEN P G, FOLEY C N, BOEKHOLT T, et al. Newton versus the machine: solving the chaotic three-body problem using deep neural networks[J]. Monthly Notices of the Royal Astronomical Society, 2020, 494(2):2465-2470. doi: 10.1093/mnras/staa713 URL
[6]	RAVURI S, LENC K, WILLSON M, et al. Skilful precipitation nowcasting using deep generative models of radar[J]. Nature, 2021, 597(7878):672-677. doi: 10.1038/s41586-021-03854-z URL
[7]	DEGRAVE J, FELICI F, BUCHLI J, et al. Magnetic control of tokamak plasmas through deep reinforcement learning[J]. Nature, 2022, 602(7897):414-419. doi: 10.1038/s41586-021-04301-9 URL
[8]	TAO F, CHENG J, QI Q, et al. Digital twin-driven product design, manufacturing and service with big data[J]. The International Journal of Advanced Manufacturing Technology, 2018, 94(9-12):3563-3576. doi: 10.1007/s00170-017-0233-1 URL
[9]	RUNCHAL A K, RAO M M. CFD of the future: year 2025 and beyond[M]. 50 Years of CFD in Engineering Sciences. Singapore: Springer Singapore, 2020:779-795.
[10]	KARNIADAKIS G E, KEVREKIDIS I G, LU L, et al. Physics-informed machine learning[J]. Nature Reviews Physics, 2021, 3(6):422-440. doi: 10.1038/s42254-021-00314-5 URL
[11]	KOCh-JANUSZ M, RINGEL Z. Mutual information, neural networks and the renormalization group[J]. Nature Physics, 2018, 14(6):578-582. doi: 10.1038/s41567-018-0081-4 URL
[12]	IULIANO E. Application of surrogate-based global optimization to aerodynamic design[M]. Springer International Publishing, 2016.
[13]	KUTZ J N. Deep learning in fluid dynamics[J]. Journal of Fluid Mechanics, 2017, 814:1-4. doi: 10.1017/jfm.2016.803 URL
[14]	THOMPSON M L, KRAMER M A. Modeling chemical processes using prior knowledge and neural networks[J]. AIChE Journal, 1994, 40(8):1328-1340. doi: 10.1002/aic.690400806 URL
[15]	LAGARIS I E, LIKAS A, FOTIADIS D I. Artificial neural networks for solving ordinary and partial differential equations[J]. IEEE Transactions on Neural Networks, 1998, 9(5):987-1000. doi: 10.1109/72.712178 pmid: 18255782
[16]	LONG Z, LU Y, DONG B. PDE-Net 2.0: learning PDEs from data with a numeric-symbolic hybrid deep network[J]. Journal of Computational Physics, 2019, 399:108925. doi: 10.1016/j.jcp.2019.108925 URL
[17]	MOHAN A T, TRETIAK D, CHERTKOV M, et al. Spatio-temporal deep learning models of 3D turbulence with physics informed diagnostics[J]. Journal of Turbulence, 2020, 21(9-10):484-524. doi: 10.1080/14685248.2020.1832230 URL
[18]	XU M, SONG S, SUN X, et al. A convolutional strategy on unstructured mesh for the adjoint vector modeling[J]. Physics of Fluids, 2021, 33(3):036115. doi: 10.1063/5.0044093 URL
[19]	KASHEFI A, REMPE D, GUIBAS L J. A point-cloud deep learning framework for prediction of fluid flow fields on irregular geometries[J]. Physics of Fluids, 2021, 33(2):027104. doi: 10.1063/5.0033376 URL
[20]	PENG W, YUAN Z, WANG J. Attention-Enhanced neural network models for turbulence simulation[J]. Physics of Fluids, 2022, 34(2):025111. doi: 10.1063/5.0079302 URL
[21]	LI Z, KOVACHKI N, AZIZZADENESHELI K, et al. Fourier neural operator for parametric partial differential equations[J]. arXiv Preprint, arXiv:2010.08895, 2020.
[22]	RAISSI M. Deep hidden physics models: deep learning of nonlinear partial differential equations[J]. arXiv Preprint, arXiv:1801.06637, 2018.
[23]	SUN L, GAO H, PAN S, et al. Surrogate modeling for fluid flows based on physics-constrained deep learning without simulation data[J]. Computer Methods in Applied Mechanics and Engineering, 2020, 361:112732. doi: 10.1016/j.cma.2019.112732 URL
[24]	RAISSI M, YAZDANI A, KARNIADAKIS G E. Hidden fluid mechanics: learning velocity and pressure fields from flow visualizations[J]. Science, 2020, 367(6481):1026-1030. doi: 10.1126/science.aaw4741 pmid: 32001523
[25]	HAGHIGHAT E, RAISSI M, MOURE A, et al. A physics-informed deep learning framework for inversion and surrogate modeling in solid mechanics[J]. Computer Methods in Applied Mechanics and Engineering, 2021, 379:113741. doi: 10.1016/j.cma.2021.113741 URL
[26]	PFAU D, SPENCER J S, MATTHEWS A, et al. Ab initio solution of the many-electron Schrdinger equation with deep neural networks[J]. Physical Review Research, 2020, 2(3).
[27]	GULIAN M, RAISSI M, PERDIKARIS P, et al. Machine learning of space-fractional differential equations[J]. SIAM Journal on Scientific Computing, 2019, 41(4):A2485-A2509. doi: 10.1137/18M1204991
[28]	RAISSI M, PERDIKARIS P, KARNIADAKIS G E. Multistep neural networks for data-driven discovery of nonlinear dynamical systems[J]. arXiv Preprint, arXiv:1801.01236, 2018.
[29]	MALINZI J, GWEBU S, MOTSA S. Determining COVID-19 dynamics using physics informed neural networks[J]. Axioms, 2022, 11(3):121. doi: 10.3390/axioms11030121 URL
[30]	LU L, MENG X, MAO Z, et al. DeepXDE: a deep learning library for solving differential equations[J]. SIAM Review, 2021, 63(1):208-228. DOI: 10.1137/19M1274067. doi: 10.1137/19M1274067 URL
[31]	CHEN F, SONDAK D, PROTOPAPAS P, et al. NeuroDiffEq: a Python package for solving differential equations with neural networks[J]. Journal of Open Source Software, 2020, 5(46):1931. doi: 10.21105/joss.01931 URL
[32]	HAGHIGHAT E, JUANES R. SciANN: a Keras/TensorFlow wrapper for scientific computations and physics-informed deep learning using artificial neural networks[J]. Computer Methods in Applied Mechanics and Engineering, 2021, 373:113552. doi: 10.1016/j.cma.2020.113552 URL
[33]	XU K, DARVE E. ADCME: learning spatially-varying physical fields using deep neural networks[J]. arXiv Preprint,arXiv:2011.11955 2020.
[34]	HENNIGH O, NARASIMHAN S, NABIAN M A, et al. NVIDIA SimNet^TM: an AI-accelerated multi-physics simulation framework[J]. arXiv Preprint,arXiv:2012.07938 2020.
[35]	NVIDIA Developer. NVIDIA modulus[EB/OL]. [2022-04-24]. https://developer.nvidia.com/modulus.
[36]	ZUBOV K, MCCarthy Z, MA Y, et al. NeuralPDE: automating physics-informed neural networks (PINNs) with error approximations[J]. arXiv Preprint,arXiv:2107.09443, 2021.
[37]	PENG W, ZHANG J, ZHOU W, et al. IDRLnet: a physics-informed neural network library[J]. arXiv Preprint, arXiv:2107.04320, 2021.
[38]	飞桨PaddlePaddle——源于产业实践的开源深度学习平台. 深度学习常用数学知识[EB/OL]. [2022-04-24]. https://www.paddlepaddle.org.cn/tutorials/projectdetail/695184.
[39]	FANG Z. A high-efficient hybrid physics-informed neural networks based on convolutional neural network[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021:1-13. DOI: 10.1109/TNNLS.2021.3070878. doi: 10.1109/TNNLS.2021.3070878
[40]	HY T S, TRIVEDI S, PAN H, et al. Predicting molecular properties with covariant compositional networks[J]. The Journal of Chemical Physics, 2018, 148(24):241745. doi: 10.1063/1.5024797 URL
[41]	LU L, JIN P, KARNIADAKIS G E. DeepONet: learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operators[J]. Nature Machine Intelligence, 2021, 3(3):218-229. doi: 10.1038/s42256-021-00302-5 URL
[42]	NIAKI S A, HAGHIGHAT E, CAMPBELL T, et al. Physics-informed neural network for modelling the thermochemical curing process of composite-tool systems during manufacture[J]. Computer Methods in Applied Mechanics and Engineering, 2021, 384(3-4):113959.
[43]	MENG X, LI Z, ZHANG D, et al. PPINN: parareal physics-informed neural network for time-dependent PDEs[J]. Computer Methods in Applied Mechanics and Engineering, 2020, 370:113250. doi: 10.1016/j.cma.2020.113250 URL
[44]	WANG S, YU X, PERDIKARIS P. When and why PINNs fail to train: a neural tangent kernel perspective[J]. arXiv Preprint, arXiv:2007.14527v1, 2020.
[45]	MOWLAVI S, NABI S. Optimal control of PDEs using physics-informed neural networks[J]. arXiv Preprint, arXiv:2111.09880, 2021.
[46]	BRUNTON S L. Machine Learning of Dynamics with Applications to Flow Control and Aerodynamic Optimization[C]. IUTAM Symposium on Critical Flow Dynamics Involving Moving/Deformable Structures with Design Applications, 2021:327-335.
[47]	CAI S, WANG Z, CHRYSSOSTOMIDIS C, et al. Heat transfer prediction with unknown thermal boundary conditions using physics-informed neural networks[C]// ASME 2020 18th International Conference on Nanochannels, Microchannels, and Minichannels[C], 2020.
[48]	KISSAS G, YANG Y, HWUANG E, et al. Machine learning in cardiovascular flows modeling: predicting arterial blood pressure from non-invasive 4D flow MRI data using physics-informed neural networks[J]. Computer Methods in Applied Mechanics and Engineering, 2020, 358:112623. doi: 10.1016/j.cma.2019.112623 URL
[49]	ZHANG E, DAO M, KARNIADAKIS G E, et al. Analyses of internal structures and defects in materials using physics-informed neural networks[J]. Science Advances, 2022, 8(7):eabk0644. doi: 10.1126/sciadv.abk0644 URL
[50]	YUCESAN Y A, VIANA F A C, MANIN L, et al. Adjusting a torsional vibration damper model with physics-informed neural networks[J]. Mechanical Systems and Signal Processing, 2021, 154:107552. doi: 10.1016/j.ymssp.2020.107552 URL
[51]	MISYRIS G S, STIASNY J, CHATZIVASILEIADIS S. Capturing power system dynamics by physics-informed neural networks and optimization[J]. arXiv Preprint, arXiv:2103.17004, 2021.
[52]	ANTONELO E A, CAMPONOGARA E, SEMAN L O, et al. Physics-informed neural nets for control of dynamical systems[J]. arXiv Preprint, arXiv:2104. 02556, 2021.
[53]	LI L, LI Y, DU Q, et al. ReF-nets: physics-informed neural network for Reynolds equation of gas bearing[J]. Computer Methods in Applied Mechanics and Engineering, 2022, 391:114524. doi: 10.1016/j.cma.2021.114524 URL
[54]	BAKARJI J, CHAMPION K, KUTZ J, et al. Discovering governing equations from partial measurements with deep delay autoencoders[J]. arXiv Preprint, arXiv:2201.05136, 2022.
[55]	CHEN W, FUGE M. BézierGAN: automatic generation of smooth curves from interpretable low-dimensional parameters[J]. arXiv Preprint, arXiv:1808.08871, 2018.
[56]	RABAULT J, REN F, ZHANG W, et al. Deep reinforcement learning in fluid mechanics: a promising method for both active flow control and shape optimization[J]. arXiv Preprint, arXiv:2001.02464, 2020.
[57]	YU L, HU W, ZHANG X, et al. Automatic generation control of ubiquitous power Internet of Things integrated energy system based on deep reinforcement learning[J]. Scientia Sinica Technologica, 2020, 50(2):221-234. doi: 10.1360/SST-2019-0199 URL
[58]	Ansys. Ansys Collaborates with GE to Drive Digital Twin Value and Deliver the Promise of the Industrial Internet of Things[EB/OL]. (2016-11-16)[2022-05-19]. https://www.ansys.com/news-center/press-releases/11-16-16-ansys-collaborates-with-ge-to-drive-digital-twin-valve-iot.
[59]	Rolls-Royce. The sustainable solution for decades to come[EB/OL]. [2022-05-19]. https://www.rolls-royce.com/media/our-stories/innovation/2016/advance-and-ultrafan.aspx.
[60]	Siemens Energy. AI makes turbines young again[EB/OL]. [2022-04-25]. https://www.siemens-energy.com/global/en/news/magazine/2020/power-station-with-real-time-rejuvenation.html.
[61]	天互数据. 百度智能云智能制造解决方案,开启中国制造的2025[EB/OL]. (2021-08-10)[2022-04-25]. https://www.idcs.cn/a/v2730.
[62]	The Defense Post. US DoD to work on project for digital copy of gas turbines[EB/OL]. (2021-08-26)[2022-04-25]. https://www.thedefensepost.com/2021/08/26/us-dod-digital-copy-gas-turbines/.

架构名称	开发语言	深度学习框架	开发机构	开发时间
DeepXDE^[30]	Python	Tensorflow	布朗大学	2020.02
NeuroDiffEq^[31]	Python	Pytorch	哈佛大学	2020.02
SciANN^[32]	Python	Keras/Tensorflow	麻省理工学院	2020.09
ADCME^[33]	Julia	Tensorflow	斯坦福大学	2020.11
SimNet^[34]/Modulus^[35]	Python	Tensorflow	英伟达公司	2020.12
NeuralPDE^[36]	Julia	Julia	卡内基梅隆大学	2021.07
IDRLnet^[37]	Python	Pytorch	中国国防科技创新研究院	2021.07
Paddle Science^[38]	Python	PaddlePaddle	百度公司	2021.12

架构名称	开发语言	深度学习框架	开发机构	开发时间
DeepXDE^[30]	Python	Tensorflow	布朗大学	2020.02
NeuroDiffEq^[31]	Python	Pytorch	哈佛大学	2020.02
SciANN^[32]	Python	Keras/Tensorflow	麻省理工学院	2020.09
ADCME^[33]	Julia	Tensorflow	斯坦福大学	2020.11
SimNet^[34]/Modulus^[35]	Python	Tensorflow	英伟达公司	2020.12
NeuralPDE^[36]	Julia	Julia	卡内基梅隆大学	2021.07
IDRLnet^[37]	Python	Pytorch	中国国防科技创新研究院	2021.07
Paddle Science^[38]	Python	PaddlePaddle	百度公司	2021.12