DiscretizationNet: A Machine-Learning based solver for Navier-Stokes Equations using Finite Volume Discretization

Over the last few decades, existing Partial Differential Equation (PDE) solvers have demonstrated a tremendous success in solving complex, non-linear PDEs. Although accurate, these PDE solvers are computationally costly. With the advances in Machine Learning (ML) technologies, there has been a significant increase in the research of using ML to solve PDEs. The goal of this work is to develop an ML-based PDE solver, that couples important characteristics of existing PDE solvers with ML technologies. The two solver characteristics that have been adopted in this work are: 1) the use of discretization-based schemes to approximate spatio-temporal partial derivatives and 2) the use of iterative algorithms to solve linearized PDEs in their discrete form. In the presence of highly non-linear, coupled PDE solutions, these strategies can be very important in achieving good accuracy, better stability and faster convergence. Our ML-solver, DiscretizationNet, employs a generative CNN-based encoder-decoder model with PDE variables as both input and output features. During training, the discretization schemes are implemented inside the computational graph to enable faster GPU computation of PDE residuals, which are used to update network weights that result into converged solutions. A novel iterative capability is implemented during the network training to improve the stability and convergence of the ML-solver. The ML-Solver is demonstrated to solve the steady, incompressible Navier-Stokes equations in 3-D for several cases such as, lid-driven cavity, flow past a cylinder and conjugate heat transfer.

[1]  Gang Bao,et al.  Weak Adversarial Networks for High-dimensional Partial Differential Equations , 2019, J. Comput. Phys..

[2]  George Em Karniadakis,et al.  A composite neural network that learns from multi-fidelity data: Application to function approximation and inverse PDE problems , 2019, J. Comput. Phys..

[3]  Balaji Srinivasan,et al.  Distributed physics informed neural network for data-efficient solution to partial differential equations , 2019, ArXiv.

[4]  George Em Karniadakis,et al.  Learning and meta-learning of stochastic advection–diffusion–reaction systems from sparse measurements , 2019, European Journal of Applied Mathematics.

[5]  Luning Sun,et al.  Surrogate modeling for fluid flows based on physics-constrained deep learning without simulation data , 2019, Computer Methods in Applied Mechanics and Engineering.

[6]  P. Perdikaris,et al.  Machine learning in cardiovascular flows modeling: Predicting arterial blood pressure from non-invasive 4D flow MRI data using physics-informed neural networks , 2019 .

[7]  Michael M. Zavlanos,et al.  VarNet: Variational Neural Networks for the Solution of Partial Differential Equations , 2019, L4DC.

[8]  Jürgen Schmidhuber,et al.  Long Short-Term Memory , 1997, Neural Computation.

[9]  George E. Karniadakis,et al.  Hidden physics models: Machine learning of nonlinear partial differential equations , 2017, J. Comput. Phys..

[10]  Stefano Ermon,et al.  Learning Neural PDE Solvers with Convergence Guarantees , 2019, ICLR.

[11]  George Em Karniadakis,et al.  NSFnets (Navier-Stokes flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations , 2020, J. Comput. Phys..

[12]  Daniel Cohen-Or,et al.  MeshCNN: a network with an edge , 2019, ACM Trans. Graph..

[13]  Paris Perdikaris,et al.  Physics-Constrained Deep Learning for High-dimensional Surrogate Modeling and Uncertainty Quantification without Labeled Data , 2019, J. Comput. Phys..

[14]  Kaj Nyström,et al.  A unified deep artificial neural network approach to partial differential equations in complex geometries , 2017, Neurocomputing.

[15]  Han Gao,et al.  PhyGeoNet: Physics-Informed Geometry-Adaptive Convolutional Neural Networks for Solving Parametric PDEs on Irregular Domain , 2020 .

[16]  Qiang Zheng,et al.  Physics-informed semantic inpainting: Application to geostatistical modeling , 2020, J. Comput. Phys..

[17]  J. Sethian,et al.  Fronts propagating with curvature-dependent speed: algorithms based on Hamilton-Jacobi formulations , 1988 .

[18]  Balaji Srinivasan,et al.  Physics Informed Extreme Learning Machine (PIELM) - A rapid method for the numerical solution of partial differential equations , 2019, Neurocomputing.

[19]  Ruben Juanes,et al.  A nonlocal physics-informed deep learning framework using the peridynamic differential operator , 2020, ArXiv.

[20]  Stephan Hoyer,et al.  Learned discretizations for passive scalar advection in a 2-D turbulent flow , 2020 .

[21]  Kaj Nyström,et al.  Data-driven discovery of PDEs in complex datasets , 2018, J. Comput. Phys..

[22]  Paris Perdikaris,et al.  Understanding and mitigating gradient pathologies in physics-informed neural networks , 2020, ArXiv.

[23]  G. Karniadakis,et al.  Physics-informed neural networks for high-speed flows , 2020, Computer Methods in Applied Mechanics and Engineering.

[24]  C. Rhie,et al.  Numerical Study of the Turbulent Flow Past an Airfoil with Trailing Edge Separation , 1983 .

[25]  Dimitrios I. Fotiadis,et al.  Artificial neural networks for solving ordinary and partial differential equations , 1997, IEEE Trans. Neural Networks.

[26]  Liu Yang,et al.  B-PINNs: Bayesian Physics-Informed Neural Networks for Forward and Inverse PDE Problems with Noisy Data , 2020, J. Comput. Phys..

[27]  George Em Karniadakis,et al.  Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations , 2020, Science.

[28]  Michael S. Triantafyllou,et al.  Deep learning of vortex-induced vibrations , 2018, Journal of Fluid Mechanics.

[29]  G. E. Karniadakis,et al.  Variational Physics-Informed Neural Networks For Solving Partial Differential Equations , 2019, ArXiv.

[30]  Zhiping Mao,et al.  DeepXDE: A Deep Learning Library for Solving Differential Equations , 2019, AAAI Spring Symposium: MLPS.

[31]  Barak A. Pearlmutter,et al.  Automatic differentiation in machine learning: a survey , 2015, J. Mach. Learn. Res..

[32]  Paris Perdikaris,et al.  Physics Informed Deep Learning (Part II): Data-driven Discovery of Nonlinear Partial Differential Equations , 2017, ArXiv.

[33]  Yan Wang,et al.  Multi-Fidelity Physics-Constrained Neural Network and Its Application in Materials Modeling , 2019, Journal of Mechanical Design.

[34]  Max Welling,et al.  Semi-Supervised Classification with Graph Convolutional Networks , 2016, ICLR.

[35]  P. Tucker,et al.  A Cartesian cut cell method for incompressible viscous flow , 2000 .

[36]  Zhiwei Fang,et al.  Deep Physical Informed Neural Networks for Metamaterial Design , 2020, IEEE Access.

[37]  G. Karniadakis,et al.  Conservative physics-informed neural networks on discrete domains for conservation laws: Applications to forward and inverse problems , 2020 .

[38]  Guofei Pang,et al.  Nonlocal Physics-Informed Neural Networks - A Unified Theoretical and Computational Framework for Nonlocal Models , 2020, AAAI Spring Symposium: MLPS.

[39]  Jason Hickey,et al.  Data-driven discretization: machine learning for coarse graining of partial differential equations , 2018 .

[40]  Hessam Babaee,et al.  Deep Learning of Turbulent Scalar Mixing , 2018, Physical Review Fluids.

[41]  A. Gelfgat,et al.  Linear instability of the lid-driven flow in a cubic cavity , 2019, Theoretical and Computational Fluid Dynamics.

[42]  George Em Karniadakis,et al.  hp-VPINNs: Variational Physics-Informed Neural Networks With Domain Decomposition , 2020, Computer Methods in Applied Mechanics and Engineering.

[43]  Hyuk Lee,et al.  Neural algorithm for solving differential equations , 1990 .

[44]  Liu Yang,et al.  Physics-Informed Generative Adversarial Networks for Stochastic Differential Equations , 2018, SIAM J. Sci. Comput..

[45]  Paris Perdikaris,et al.  Learning Parameters and Constitutive Relationships with Physics Informed Deep Neural Networks , 2018, 1808.03398.

[46]  Jung Hee Seo,et al.  A sharp-interface immersed boundary method with improved mass conservation and reduced spurious pressure oscillations , 2011, J. Comput. Phys..

[47]  Ke Li,et al.  D3M: A Deep Domain Decomposition Method for Partial Differential Equations , 2020, IEEE Access.

[48]  Suhas V. Patankar,et al.  A Calculation Procedure for Two-Dimensional Elliptic Situations , 1981 .

[49]  Tim Colonius,et al.  FiniteNet: A Fully Convolutional LSTM Network Architecture for Time-Dependent Partial Differential Equations , 2020, ArXiv.

[50]  George Em Karniadakis,et al.  On the Convergence and generalization of Physics Informed Neural Networks , 2020, ArXiv.

[51]  L. Dal Negro,et al.  Physics-informed neural networks for inverse problems in nano-optics and metamaterials. , 2019, Optics express.

[52]  George Em Karniadakis,et al.  PPINN: Parareal Physics-Informed Neural Network for time-dependent PDEs , 2019, Computer Methods in Applied Mechanics and Engineering.