Numerical simulations using the Discrete Element Method (DEM) are used at LEAT in the context of several important, energy related particulate flow systems. The focus of these investigations is the understanding of the heat and mass transfer processes on the micro-scale and the prediction of the related macroscopic behaviour. Most of the currently available DEM implementations, especially if the required number of particles is large, only allow for small variations in particle size if the computational effort must be kept within reasonable bounds. This is contrary to the actual requirements of many technically relevant processes where a broad size spectrum must be considered. Parallel processing helps to ease this situation to a certain degree, but the ongoing search for algorithmic improvements has not yet accomplished a definitive solution.
The process of neighbourhood detection, which is required to identify the partners of the pairwise interactions determining momentum fluxes among the particles and between particles and surrounding walls is one common bottleneck. Besides the commonly used Linked-Cell method, hierarchically structured “background” meshes or octrees were proposed in the past and applied in various implementations. A new variant of the octree approach is presented and its performance with respect to particle number, particle size distribution and parallelisation is analysed and compared to conventional approaches. In order to obtain a realistic analysis, for a given code in a typical hardware environment (small engineering companies or university institutes), the benchmark addresses the technical application of particle movement in a section of a rotary drum.