Abstract:The resolution of the Poisson equation is usually one of the most computationally intensive steps for incompressible fluid solvers. Lately, Deep Learning, and especially Convolutional Neural Networks (CNN), has been introduced to solve this equation, leading to significant inference time reduction at the cost of a lack of guarantee on the accuracy of the solution. This drawback might lead to inaccuracies and potentially unstable simulations. It also makes impossible a fair assessment of the CNN speedup, for instance, when changing the network architecture, since evaluated at different error levels. To circumvent this issue, a hybrid strategy is developed, which couples a CNN with a traditional iterative solver to ensure a user-defined accuracy level. The CNN hybrid method is tested on two flow cases, consisting of a variable-density plume with and without obstacles, demostrating remarkable generalization capabilities, ensuring both the accuracy and stability of the simulations. The error distribution of the predictions using several network architectures is further investigated. Results show that the threshold of the hybrid strategy defined as the mean divergence of the velocity field is ensuring a consistent physical behavior of the CNN-based hybrid computational strategy. This strategy allows a systematic evaluation of the CNN performance at the same accuracy level for various network architectures. In particular, the importance of incorporating multiple scales in the network architecture is demonstrated, since improving both the accuracy and the inference performance compared with feedforward CNN architectures, as these networks can provide solutions 1 10-25 faster than traditional iterative solvers.

Construction of Elliptic Stochastic Partial Differential Equations Solver in Groundwater Flow with Convolutional Neural Networks

A Deep Neural Network Method for Solving Partial Differential Equations with Complex Boundary in Groundwater Seepage

Learning Time-Dependent PDEs with a Linear and Nonlinear Separate Convolutional Neural Network

Deep Latent Regularity Network for Modeling Stochastic Partial Differential Equations

Solving Inverse Problems in Steady State Navier-Stokes Equations using Deep Neural Networks

A Novel CNN-Based Poisson Solver for Fluid Simulation

A Non-Gradient Method for Solving Elliptic Partial Differential Equations with Deep Neural Networks.

A fully differentiable GNN-based PDE Solver: With Applications to Poisson and Navier-Stokes Equations

Deep FBSDE Neural Networks for Solving Incompressible Navier-Stokes Equation and Cahn-Hilliard Equation

Solution of diffusivity equations with local sources/sinks and surrogate modeling using weak form theory-guided neural network

Numerical approximation based on deep convolutional neural network for high‐dimensional fully nonlinear merged PDEs and 2BSDEs

Hybrid Neural Networks for Solving Fully Coupled, High-Dimensional Forward–Backward Stochastic Differential Equations

Convolution-Based Model-Solving Method for Three-Dimensional, Unsteady, Partial Differential Equations

PhyGeoNet: Physics-informed Geometry-Adaptive Convolutional Neural Networks for Solving Parameterized Steady-State PDEs on Irregular Domain

Neural-PDE: a RNN Based Neural Network for Solving Time Dependent PDEs

Study on a Poisson's Equation Solver Based on Deep Learning Technique.

Performance and accuracy assessments of an incompressible fluid solver coupled with a deep Convolutional Neural Network

Time integration schemes based on neural networks for solving partial differential equations on coarse grids

Theory-Guided Convolutional Neural Network with an Enhanced Water Flow Optimizer.

Solving stochastic partial differential equations using neural networks in the Wiener chaos expansion

Solve High-Dimensional Reflected Partial Differential Equations by Neural Network Method