Journal article 294 views 7 downloads
Meshing using neural networks for improving the efficiency of computer modelling
Engineering with Computers
PDF | Version of Record
© The Author(s) 2023. Distributed under the terms of a Creative Commons Attribution 4.0 License (CC BY 4.0).Download (6.57MB)
This work presents a novel approach capable of predicting an appropriate spacing function that can be used to generate a near-optimal mesh suitable for simulation. The main objective is to make use of the large number of simulations that are nowadays available, and to alleviate the time-consuming me...
|Published in:||Engineering with Computers|
Springer Science and Business Media LLC
Check full text
No Tags, Be the first to tag this record!
This work presents a novel approach capable of predicting an appropriate spacing function that can be used to generate a near-optimal mesh suitable for simulation. The main objective is to make use of the large number of simulations that are nowadays available, and to alleviate the time-consuming mesh generation stage by minimising human intervention. For a given simulation, a technique to produce a set of point sources that leads to a mesh capable of capturing all the features of the solution is proposed. In addition, a method to combine all sets of sources for the simulations available is devised. The global set of sources is used to train a neural network that, for some design parameters (e.g., flow conditions, geometry), predicts the characteristics of the sources. Numerical examples, in the context of three dimensional inviscid compressible flows, are considered to demonstrate the potential of the proposed approach. It is shown that accurate predictions of the required spacing function can be produced, even with reduced training datasets. In addition, the predicted near-optimal meshes are utilised to compute flow solutions, and the results show that the computed aerodynamic coefficients are within the required accuracy for the aerospace industry. An analysis is also presented to demonstrate that the proposed method lies in the category of green AI research, meaning that computational resources and time are substantially reduced with this approach, when compared to current practice in industry.
Mesh generation, Spacing function, Machine learning, Near-optimal mesh prediction, Computational fluid dynamics
Faculty of Science and Engineering
Swansea University. The authors are grateful for the financial support provided by the Engineering and Physical Sciences Research Council (EP/T517987/1).