Smart Adaptive Mesh Refinement with NEMoSys

TL;DR

This paper presents a novel smart adaptive mesh refinement method that combines classical AMR with machine learning, aiming to improve efficiency and address limitations of traditional approaches in computational simulations.

Contribution

The paper introduces a new machine learning-based approach for adaptive mesh refinement, including an algorithm, modular structure, and training procedures, demonstrated on CFD problems.

Findings

Feasibility of smart AMR shown on CFD problems

Preliminary numerical results indicate improved efficiency

Modular object-oriented implementation developed

Abstract

Adaptive mesh refinement (AMR) offers a practical solution to reduce the computational cost and memory requirement of numerical simulations that use computational meshes. In this work, we introduce a novel smart methodology for adaptive mesh refinement. Smart adaptive refinement blends classical AMR with machine learning to address some of the known issues of the conventional approaches. We provide an algorithm for adaptive refinement. Subsequently, we introduce a modular object-oriented structure for our smart AMR algorithm. Then we present procedures used for the training of a smart AMR model. The study follows with a demonstration of preliminary numerical studies indicating the feasibility of performing adaptive mesh refinement on a few demonstrative problems selected from the CFD domain. Finally, we conclude with a few comments about future work.