Physics-informed ConvNet: learning physical field from a shallow neural network

DOI10.1016/J.CNSNS.2024.107911arXiv2201.10967OpenAlexW4391850501MaRDI QIDQ6199712FDOQ6199712

Authors: Pengpeng Shi, Zhi Zeng, Tianshou Liang 

Publication date: 21 March 2024

Published in: Communications in Nonlinear Science and Numerical Simulation (Search for Journal in Brave)

Abstract: Big-data-based artificial intelligence (AI) supports profound evolution in almost all of science and technology. However, modeling and forecasting multi-physical systems remain a challenge due to unavoidable data scarcity and noise. Improving the generalization ability of neural networks by "teaching" domain knowledge and developing a new generation of models combined with the physical laws have become promising areas of machine learning research. Different from "deep" fully-connected neural networks embedded with physical information (PINN), a novel shallow framework named physics-informed convolutional network (PICN) is recommended from a CNN perspective, in which the physical field is generated by a deconvolution layer and a single convolution layer. The difference fields forming the physical operator are constructed using the pre-trained shallow convolution layer. An efficient linear interpolation network calculates the loss function involving boundary conditions and the physical constraints in irregular geometry domains. The effectiveness of the current development is illustrated through some numerical cases involving the solving (and estimation) of nonlinear physical operator equations and recovering physical information from noisy observations. Its potential advantage in approximating physical fields with multi-frequency components indicates that PICN may become an alternative neural network solver in physics-informed machine learning.

Full work available at URL: https://arxiv.org/abs/2201.10967

Recommendations

• f-PICNN: a physics-informed convolutional neural network for partial differential equations with space-time domain
• Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
• PhyGeoNet: physics-informed geometry-adaptive convolutional neural networks for solving parameterized steady-state PDEs on irregular domain
• PhyCRNet: physics-informed convolutional-recurrent network for solving spatiotemporal PDEs
• Physics-informed neural networks combined with polynomial interpolation to solve nonlinear partial differential equations

zbMATH Keywords

machine learningdata-driven scientific computingdenoising displaymulti-frequency fieldoperator estimatingphysics-informed ConvNet(PICN)

Mathematics Subject Classification ID

Learning and adaptive systems in artificial intelligence (68T05) Large deviations (60F10) Numerical interpolation (65D05) Nonlinear ordinary differential equations and systems (34A34) Time-dependent Schrödinger equations and Dirac equations (35Q41) NLS equations (nonlinear Schrödinger equations) (35Q55) Series expansions (e.g., Taylor, Lidstone series, but not Fourier series) (41A58) Finite difference methods for initial value and initial-boundary value problems involving PDEs (65M06) Stability and convergence of numerical methods for initial value and initial-boundary value problems involving PDEs (65M12) Finite difference methods for boundary value problems involving PDEs (65N06) PDEs in connection with classical thermodynamics and heat transfer (35Q79) PDEs on graphs and networks (ramified or polygonal spaces) (35R02) Numerical methods for partial differential equations, initial value and time-dependent initial-boundary value problems (65M99)

Cites Work

• Machine learning: Trends, perspectives, and prospects
• Title not available (Why is that?)
• A Multigrid Tutorial, Second Edition
• Solution of nonlinear ordinary differential equations by feedforward neural networks
• Neural algorithm for solving differential equations
• Neural‐network‐based approximations for solving partial differential equations
• Physics-constrained deep learning for high-dimensional surrogate modeling and uncertainty quantification without labeled data
• Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
• Deep hidden physics models: deep learning of nonlinear partial differential equations
• ConvPDE-UQ: convolutional neural networks with quantified uncertainty for heterogeneous elliptic partial differential equations on varied domains
• PhyGeoNet: physics-informed geometry-adaptive convolutional neural networks for solving parameterized steady-state PDEs on irregular domain
• Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations
• Lift \& learn: physics-informed machine learning for large-scale nonlinear dynamical systems
• Deep learning solver for solving advection-diffusion equation in comparison to finite difference methods
• PhyCRNet: physics-informed convolutional-recurrent network for solving spatiotemporal PDEs
• Modeling the dynamics of PDE systems with physics-constrained deep auto-regressive networks
• On the eigenvector bias of Fourier feature networks: from regression to solving multi-scale PDEs with physics-informed neural networks
• Parallel physics-informed neural networks via domain decomposition
• Scientific machine learning through physics-informed neural networks: where we are and what's next
• Shallow neural networks for fluid flow reconstruction with limited sensors
• Physics-informed neural networks with hard constraints for inverse design
• Conditional physics informed neural networks
• A study on data-driven identification and representation of nonlinear dynamical systems with a physics-integrated deep learning approach: Koopman operators and nonlinear normal modes
• Three ways to solve partial differential equations with neural networks — A review
• Data-driven forward-inverse problems for Yajima-Oikawa system using deep learning with parameter regularization
• Dynamic analysis on optical pulses via modified PINNs: soliton solutions, rogue waves and parameter discovery of the CQ-NLSE

Cited In (10)

• Machine learning for nonlinear integro-differential equations with degenerate kernel scheme
• Pseudo grid-based physics-informed convolutional-recurrent network solving the integrable nonlinear lattice equations
• f-PICNN: a physics-informed convolutional neural network for partial differential equations with space-time domain
• Filtered partial differential equations: a robust surrogate constraint in physics-informed deep learning framework
• nPINNs: nonlocal physics-informed neural networks for a parametrized nonlocal universal Laplacian operator. Algorithms and applications
• Physics-informed radial basis network (PIRBN): a local approximating neural network for solving nonlinear partial differential equations
• Prediction of mechanical property of open-hole composite laminates using generalized regression neural network method
• Navigating PINNs via maximum residual-based continuous distribution
• Physics informed WNO
• Solving seepage equation using physics-informed residual network without labeled data

This page was built for publication: Physics-informed ConvNet: learning physical field from a shallow neural network

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6199712)