The Poisson equation is commonly encountered in engineering, for instance, in computational fluid dynamics (CFD) where it is needed to compute corrections to the pressure field to ensure the incompressibility of the velocity field. In the present work, we propose a novel fully convolutional neural network (CNN) architecture to infer the solution of the Poisson equation on a 2D Cartesian grid with different resolutions given the right-hand side term, arbitrary boundary conditions, and grid parameters. It provides unprecedented versatility for a CNN approach dealing with partial differential equations. The boundary conditions are handled using a novel approach by decomposing the original Poisson problem into a homogeneous Poisson problem plus four inhomogeneous Laplace subproblems. The model is trained using a novel loss function approximating the continuous $ {L}^p $ norm between the prediction and the target. Even when predicting on grids denser than previously encountered, our model demonstrates encouraging capacity to reproduce the correct solution profile. The proposed model, which outperforms well-known neural network models, can be included in a CFD solver to help with solving the Poisson equation. Analytical test cases indicate that our CNN architecture is capable of predicting the correct solution of a Poisson problem with mean percentage errors below 10%, an improvement by comparison to the first step of conventional iterative methods. Predictions from our model, used as the initial guess to iterative algorithms like Multigrid, can reduce the root mean square error after a single iteration by more than 90% compared to a zero initial guess.
17 June 2020 in International Journal for Numerical Methods in Fluids
Summary: High‐speed compressible turbulent flows typically contain discontinuities and have been widely modeled using Weighted Essentially Non‐Oscillatory (WENO) schemes due to their high‐order accuracy and sharp shock capturing capability. However, such schemes may damp the small scales of turbulence and result in inaccurate solutions in the context of turbulence‐resolving simulations. In this connection, the recently developed Targeted Essentially Non‐Oscillatory (TENO) schemes, including adaptive variants, may offer significant improvements. The present study aims to quantify the potential of these new schemes for a fully turbulent supersonic flow. Specifically, DNS of a compressible turbulent channel flow with M = 1.5 and Reτ = 222 is conducted using OpenSBLI, a high‐order finite difference computational fluid dynamics framework. This flow configuration is chosen to decouple the effect of flow discontinuities and turbulence and focus on the capability of the aforementioned high‐order schemes to resolve turbulent structures. The effect of the spatial resolution in different directions and coarse grid implicit LES are also evaluated against the WALE LES model. The TENO schemes are found to exhibit significant performance improvements over the WENO schemes in terms of the accuracy of the statistics and the resolution of the three‐dimensional vortical structures. The sixth‐order adaptive TENO scheme is found to produce comparable results to those obtained with nondissipative fourth‐ and sixth‐order central schemes and reference data obtained with spectral methods. Although the most computationally expensive scheme, it is shown that this adaptive scheme can produce satisfactory results if used as an implicit LES model.
Direct numerical simulations of two-dimensional (2D) and three-dimensional (3D), single-mode and multi-mode, incompressible immiscible Rayleigh–Taylor (RT) instabilities are performed using a phase-field approach and high-order finite-difference schemes. Various combinations of Atwood number, Reynolds number, surface tension, and initial perturbation amplitude are investigated. It is found that at high Reynolds numbers, the surface tension, if significant, could prevent the formation of Kelvin–Helmholtz type instabilities within the bubble region. A relationship is proposed for the vertical distance of the bubble and spike vs the Atwood number. The spike and bubble reaccelerate after reaching a temporary plateau due to the reduction of the friction drag as a result of the formation of the spike vortices and also the formation of a momentum jet traveling upward within the bubble region. The interface for a 3D single-mode instability grows exponentially; however, a higher Reynolds number and/or a lower Atwood number could result in a noticeably larger surface area after the initial growth. It is also shown that a 3D multi-mode RT instability initially displays an exponential interface growth rate similar to single-mode RT instabilities. Due to the collapse and merging of individual single-mode instabilities, the interface area for a multi-mode RT instability is strongly dependent to the mesh resolution after the exponential growth rate. However, the ratio of kinetic energy over released potential energy exhibits an almost steady state after the initial exponential growth, with values around 0.4, independently of the mesh resolution.
11 October 2023 in International Journal of Heat and Fluid Flow
Accurate and efficient modeling of atmospheric composition, including aerosols and trace gases and their interactions with radiation, clouds, and dynamics is essential for improving predictions and understanding of air quality, weather, climate, and related health impacts. The Aerosols and Reactive Trace gases (ART) component extends the ICOsahedral Nonhydrostatic (ICON) modeling framework by enabling online, fully coupled simulations of atmospheric composition processes across scales. ICON-ART includes modules for emissions, transport, gas-phase chemistry, and aerosol microphysics in both the troposphere and stratosphere, allowing for the investigation of feedbacks between atmospheric composition and physical processes from the large-eddy to global scale. This paper presents an updated overview of the ICON-ART framework as implemented in version 2025.10, highlighting recent developments in emission parameterizations, chemical mechanisms, aerosol processes, and coupling to the physical core of ICON via aerosol–radiation and aerosol–cloud interactions. We summarize the structure of the code infrastructure and demonstrate the model’s flexibility and scalability across a wide range of applications. ICON-ART provides a unified and modular platform for research and operational use in atmospheric composition, bridging the gap between regional air quality modeling and global Earth system simulations.