Deep convolutional Ritz method: parametric PDE surrogates without labeled data

DOI10.1007/S10483-023-2992-6arXiv2206.04675OpenAlexW4382894964MaRDI QIDQ6132294FDOQ6132294

Authors: Jan N. Fuhg, Ashwini Karmarkar, Teeratorn Kadeethum, H. Yoon, Nikolaos Bouklas 

Publication date: 14 July 2023

Published in: AMM. Applied Mathematics and Mechanics. (English Edition) (Search for Journal in Brave)

Abstract: Parametric surrogate models for partial differential equations (PDEs) are a necessary component for many applications in the computational sciences, and convolutional neural networks (CNNs) have proved as an excellent tool to generate these surrogates when parametric fields are present. CNNs are commonly trained on labeled data based on one-to-one sets of parameter-input and PDE-output fields. Recently, residual-based convolutional physics-informed neural network (CPINN) solvers for parametric PDEs have been proposed to build surrogates without the need for labeled data. These allow for the generation of surrogates without an expensive offline-phase. In this work, we present an alternative formulation termed Deep Convolutional Ritz Method (DCRM) as a parametric PDE solver. The approach is based on the minimization of energy functionals, which lowers the order of the differential operators compared to residual-based methods. Based on studies involving the Poisson equation with a spatially parameterized source term and boundary conditions, we found that CNNs trained on labeled data outperform CPINNs in convergence speed and generalization ability. Surrogates generated from DCRM, however, converge significantly faster than their CPINN counterparts and prove to generalize faster and better than surrogates obtained from both CNNs trained on labeled data and CPINNs. This hints that DCRM could make PDE solution surrogates trained without labeled data possible.

Full work available at URL: https://arxiv.org/abs/2206.04675

zbMATH Keywords

convolutional neural network (CNN)deep energy networkphysics-informed constraint

Mathematics Subject Classification ID

Neural networks for/in biological studies, artificial life and related topics (92B20) Numerical analysis (65-XX) Partial differential equations (35-XX)

Cites Work

• Title not available (Why is that?)
• Reducing the Dimensionality of Data with Neural Networks
• Title not available (Why is that?)
• Title not available (Why is that?)
• Title not available (Why is that?)
• Calibrated reduced-order POD-Galerkin system for fluid flow modelling
• Progressive construction of a parametric reduced‐order model for PDE‐constrained optimization
• Nonlinear finite element methods
• The Hitchhiker's Guide to the Virtual Element Method
• A global version of the inverse problem of the calculus of variations
• On the Existence of Global Variational Principles
• A reduced multiscale model for nonlinear structural topology optimization
• Title not available (Why is that?)
• Title not available (Why is that?)
• Title not available (Why is that?)
• Model reduction and neural networks for parametric PDEs
• Solving parametric PDE problems with artificial neural networks
• POLYNOMIAL-CHAOS-BASED KRIGING
• Large-scale PDE-constrained optimization
• The finite volume method in computational fluid dynamics. An advanced introduction with OpenFOAM and Matlab
• Model-data-driven constitutive responses: application to a multiscale computational framework
• Bayesian deep convolutional encoder-decoder networks for surrogate modeling and uncertainty quantification
• The Deep Ritz Method: a deep learning-based numerical algorithm for solving variational problems
• 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
• Reduced Basis Methods for Uncertainty Quantification
• Reduced order modeling for nonlinear structural analysis using Gaussian process regression
• Deep UQ: learning deep neural network surrogate models for high dimensional uncertainty quantification
• The Inverse Problem of the Calculus of Variations
• PhyGeoNet: physics-informed geometry-adaptive convolutional neural networks for solving parameterized steady-state PDEs on irregular domain
• When and why PINNs fail to train: a neural tangent kernel perspective
• A Gaussian process regression approach within a data-driven POD framework for engineering problems in fluid dynamics
• Improving the efficiency of large scale topology optimization through on-the-fly reduced order model construction
• An energy approach to the solution of partial differential equations in computational mechanics via machine learning: concepts, implementation and applications
• The neural particle method - an updated Lagrangian physics informed neural network for computational fluid dynamics
• Local approximate Gaussian process regression for data-driven constitutive models: development and comparison with neural networks
• A theoretical analysis of deep neural networks and parametric PDEs
• PhyCRNet: physics-informed convolutional-recurrent network for solving spatiotemporal PDEs
• Deep Nitsche Method: Deep Ritz Method with Essential Boundary Conditions
• Robust topology optimization with low rank approximation using artificial neural networks
• The mixed deep energy method for resolving concentration features in finite strain hyperelasticity
• Convergence Rate Analysis for Deep Ritz Method
• On physics-informed data-driven isotropic and anisotropic constitutive models through probabilistic machine learning and space-filling sampling
• Title not available (Why is that?)
• A heteroencoder architecture for prediction of failure locations in porous metals using variational inference
• Solution of the Inverse Problem of the Calculus of Variations
• A physics-informed neural network technique based on a modified loss function for computational 2D and 3D solid mechanics

Cited In (1)

• Preface: machine-learning approaches for computational mechanics

This page was built for publication: Deep convolutional Ritz method: parametric PDE surrogates without labeled data

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6132294)