I Simposio de Postgrado 2023. Ingeniería, ciencias e innovación

MÓDULO_ 03 Computación y Ciencias de datos 74 POLYLLA: POLYGONAL MESHING ALGORITHM BASED ON TERMINAL-EDGE REGIONS ABSTRACT Polygonal/Polyhedral mesh generation is becoming increasingly important as innovative numerical methods rely onmeshes to discretize domains. Finite Element Method (FEM) simulations use triangular and quadrangular meshes for various applications such as heat transfer, fluid dynamics, and fracture mechanics. However, ensuring that FEM elements meet quality criteria often requires inserting numerous points and elements, resulting in longer simulations. The Virtual Element Method (VEM) is promising because it can use any polygon as a basic cell, even those with arbitrary shapes, reducing the number of cells andmesh points to improve simulation speed. However, automatic algorithms for arbitrary polygon shapes in VEM are lacking. This work introduces Polylla , an innovative polygonal mesh generator engineered to adapt seamlessly to any geometric domain. The research showcases outstanding achievements that distinguishPolylla fromthe current stateof the art inmeshing tools: The Polylla meshing tool generates meshes 573 times faster than the CGALVoronoimesh generator. Polylla meshes show a similar computational cost in time and accuracy to Voronoimeshes in experiments performed with the VEM. The parallel GPU version has managed to accelerate the time of generation of Polylla meshes 90 times faster compared to its sequential version. The compact version of Polylla allows us to compress 99% the memory used by the Polylla meshes and reduce the memory cost necessary for the generation ofthemeshesbyafactor of 3. The study results under score the remark able effectiveness of Polylla in rapidly generating high-quality polygonal meshes of arbitrary size for representing complex real-life geometries. Polylla’s potential to revolutionize mesh generation techniques across diverse scientific and engineering applications is evident. A gallery of Polylla meshes is provided for reference. 1 DCC,Universidad de Chile. *Email: sergiosalinas@uchile.cl Sergio Salinas 1* , Nancy Hitschfeld 1