Active training of physics-informed neural networks to aggregate and interpolate parametric solutions to the Navier-Stokes equations

被引:38
作者
Arthurs, Christopher J. [1 ]
King, Andrew P. [1 ]
机构
[1] Kings Coll London, St Thomas Hosp, Sch Biomed Engn & Imaging Sci, 4th Floor Lambeth Wing, London SE1 7EH, England
基金
英国惠康基金; 英国工程与自然科学研究理事会;
关键词
Navier-Stokes; Deep learning; Active learning; Training; Computing methods; SIMULATION; FLOW;
D O I
10.1016/j.jcp.2021.110364
中图分类号
TP39 [计算机的应用];
学科分类号
080201 [机械制造及其自动化];
摘要
The goal of this work is to train a neural network which approximates solutions to the Navier-Stokes equations across a region of parameter space, in which the parameters define physical properties such as domain shape and boundary conditions. The contributions of this work are threefold: 1. To demonstrate that neural networks can be efficient aggregators of whole families of parametric solutions to physical problems, trained using data created with traditional, trusted numerical methods such as finite elements. Advantages include extremely fast evaluation of pressure and velocity at any point in physical and parameter space (asymptotically, similar to 3 mu s/query), and data compression (the network requires 99% less storage space compared to its own training data). 2. To demonstrate that the neural networks can accurately interpolate between finite element solutions in parameter space, allowing them to be instantly queried for pressure and velocity field solutions to problems for which traditional simulations have never been performed. 3. To introduce an active learning algorithm, so that during training, a finite element solver can automatically be queried to obtain additional training data in locations where the neural network's predictions are in most need of improvement, thus autonomously acquiring and efficiently distributing training data throughout parameter space. In addition to the obvious utility of Item 2, above, we demonstrate an application of the network in rapid parameter sweeping, very precisely predicting the degree of narrowing in a tube which would result in a 50% increase in end-to-end pressure difference at a given flow rate. This capability could have applications in both medical diagnosis of arterial disease, and in computer-aided design. (C) 2021 The Author(s). Published by Elsevier Inc.
引用
收藏
页数:17
相关论文
共 33 条
[1]  
Abadi M, 2016, PROCEEDINGS OF OSDI'16: 12TH USENIX SYMPOSIUM ON OPERATING SYSTEMS DESIGN AND IMPLEMENTATION, P265
[2]   Development of a patient-specific simulation tool to analyse aortic dissections: Assessment of mixed patient-specific flow and pressure boundary conditions [J].
Alimohammadi, Mona ;
Agu, Obiekezie ;
Balabani, Stavroula ;
Diaz-Zuccarini, Vanessa .
MEDICAL ENGINEERING & PHYSICS, 2014, 36 (03) :275-284
[3]  
[Anonymous], 2009, COMPUTER SCI TECHNIC
[4]  
Arthurs C.J., 2020, 14 WORLD C COMP MECH
[5]   Reproducing Patient-Specific Hemodynamics in the Blalock-Taussig Circulation Using a Flexible Multi-Domain Simulation Framework: Applications for Optimal Shunt Design [J].
Arthurs, Christopher J. ;
Agarwal, Pradyumn ;
John, Anna V. ;
Dorfman, Adam L. ;
Grifka, Ronald G. ;
Figueroa, C. Alberto .
FRONTIERS IN PEDIATRICS, 2017, 5
[6]   Efficient simulation of cardiac electrical propagation using high-order finite elements II: Adaptive p-version [J].
Arthurs, Christopher J. ;
Bishop, Martin J. ;
Kay, David .
JOURNAL OF COMPUTATIONAL PHYSICS, 2013, 253 :443-470
[7]   Efficient simulation of cardiac electrical propagation using high order finite elements [J].
Arthurs, Christopher J. ;
Bishop, Martin J. ;
Kay, David .
JOURNAL OF COMPUTATIONAL PHYSICS, 2012, 231 (10) :3946-3962
[8]   Nektar plus plus : An open-source spectral/hp element framework [J].
Cantwell, C. D. ;
Moxey, D. ;
Comerford, A. ;
Bolis, A. ;
Rocco, G. ;
Mengaldo, G. ;
De Grazia, D. ;
Yakovlev, S. ;
Lombard, J. -E. ;
Ekelschot, D. ;
Jordi, B. ;
Xu, H. ;
Mohamied, Y. ;
Eskilsson, C. ;
Nelson, B. ;
Vos, P. ;
Biotto, C. ;
Kirby, R. M. ;
Sherwin, S. J. .
COMPUTER PHYSICS COMMUNICATIONS, 2015, 192 :205-219
[9]   Physics-informed neural networks for inverse problems in nano-optics and metamaterials [J].
Chen, Yuyao ;
Lu, Lu ;
Karniadakis, George Em ;
Dal Negro, Luca .
OPTICS EXPRESS, 2020, 28 (08) :11618-11633
[10]   Gmsh: A 3-D finite element mesh generator with built-in pre- and post-processing facilities [J].
Geuzaine, Christophe ;
Remacle, Jean-Francois .
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING, 2009, 79 (11) :1309-1331