A Neural Network-Based Mesh Quality Indicator for Three-Dimensional Cylinder Modelling.

Area of Science:

• Computational Fluid Dynamics (CFD)
• Numerical Simulation
• Mesh Generation

Background:

• Traditional mesh quality indicators for CFD simulations are often insufficient, relying solely on geometric information of individual elements.
• Manual re-evaluation of mesh properties like distribution and refinement is time-consuming, increasing meshing overhead.
• Accurate mesh quality evaluation is crucial for reliable cylinder modeling in CFD.

Purpose of the Study:

• To develop an efficient and automated quality indicator for varisized cylinder meshes in CFD simulations.
• To reduce the manual effort and overhead associated with traditional mesh quality assessment.
• To improve the accuracy of CFD simulations by considering comprehensive mesh quality properties.

Main Methods:

• Introduction of Mesh-Net, a novel indicator combining a mesh pre-processing method and a neural network.
• Training Mesh-Net on a published cylinder mesh benchmark dataset to learn the impact of mesh quality on simulation accuracy.
• Incorporating both element geometry (e.g., orthogonality) and mesh quality properties (e.g., smoothness, distribution) into the indicator.

Main Results:

• The proposed Mesh-Net indicator accurately predicts the overall quality of input meshes.
• Experimental results validate the effectiveness of the neural network-based approach.
• The indicator successfully automates mesh quality evaluation without requiring manual interaction.

Conclusions:

• Mesh-Net provides an accurate and efficient solution for evaluating cylinder mesh quality in CFD.
• The automated approach significantly reduces the overhead associated with manual mesh assessment.
• This method enhances the reliability and accuracy of CFD simulations by ensuring high-quality meshes.