Particle packing strategies for modelling granular media

• Ano de defesa: 2019
• Autor(a) principal: Lopes, Lucas Gouveia Omena
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso embargado
• Idioma: por
• Instituição de defesa: Universidade Federal de Alagoas; Brasil; Programa de Pós-Graduação em Engenharia Civil; UFAL
• Palavras-chave em Português: Modelagem de meios granulares; Geração de partículas; Método dos elementos discretos; Otimização; Computação de alto desempenho; Granular Media Modeling; Particle Generation; Discrete Element Method; Optimization; High Performace Computing; CNPQ::ENGENHARIAS::ENGENHARIA CIVIL
• Link de acesso: http://www.repositorio.ufal.br/handle/riufal/6091

This work presents strategies for modeling numerical granular media that reproduce characteristics present in real particulate aggregates, such as soils. The presented methodologies have the objective to achieve homogeneous granular models respecting prescribed filling rate and grain size distributions. The strategies can be used by simulators of particulate media to create the initial particle arrangement, such as those used in the Discrete Element Method (DEM). Examples of DEM applications are numerical simulations of fragmentation, fracturing, impact and collision phenomena, as well as those involving soils and rocks as integral media of geomechanical problems. The literature divides techniques for generating particulate media into two categories: geometric and dynamic procedures three alternative methodologies to those present in the literature are explored in this work: i) distance minimization; ii) serial geometric separation technique; and iii) parallel alternative geometric separation technique. The presented approaches deal with geometric procedures starting from an input model with particles in random positions. The overlapping between the particles is eliminated during the proposed methodology procedures, and an output model is delivered. An important contribution of this work is the development of techniques that improve the processing time for modeling using parallel implementation with a focus on GPUs (Graphics Processing Units). Granular models generated from real soil data are presented for validation of the proposed strategies. The results show that the techniques are capable of generating granular models with characteristics close to those of real aggregates. The use of parallel strategies in GPU also presents significant gains, reaching speedups above 200.