The Physalis Method for Particulate Flow Simulations: Applications and Extensions

computational fluid dynamics;multiphase flows;particulate flows;Mechanical Engineering

Willen, Daniel PProsperetti, Andrea ;

Johns Hopkins University

关键词: computational fluid dynamics;    multiphase flows;    particulate flows;    Mechanical Engineering;   

Others  :  https://jscholarship.library.jhu.edu/bitstream/handle/1774.2/60061/WILLEN-DISSERTATION-2018.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: JOHNS HOPKINS DSpace Repository

【 摘 要 】

Particulate multiphase flows, in which a disperse solid phase is immersed in a continuous fluid medium, are relevant in a wide variety of natural and technological phenomena. Examples include dust storms, avalanches, and sediment transport, as well as fluidized bed reactors, grain transport, and combustion. The prediction and understanding of such flows is highly desirable, but the large range of scales involved presents a difficult challenge for researchers. For instance, dust storms can be several kilometers long but contain particles whose diameter is on the order of tens of micrometers. This makes it difficult to account for all the scales in experimental or numerical investigations and so much recent work has focused on investigating highly resolved simulations of small-scale flows with the intent of gaining insight into the microscopic interactions between the fluid and the particles. This new knowledge can then be applied to reduced-order models capable of capturing macroscopic phenomena. In the present work, we report on several such investigations of particulate flows, with a focus on a set of simulations representing triply periodic sedimentation of a suspension composed of solid spherical particles immersed in a fluid. Using a technique to represent the disperse particle phase as a (coarse-grained) continuous field, we identify the presence of continuity waves in the simulations and measure their wave speed. We find excellent agreement of the measured wave speed with a prediction based on a one-dimensional balance equation for the particle volume fraction, which indicates that the method used to coarse-grain the particle phase accurately captures the macroscopic features of the flow.Additionally, we investigate elements of the suspension microstructure concerning particle collisions, diffusivities, mean free paths, pair distribution function and other features. It is found that many qualitative trends found in earlier studies continue to hold in the parameter range investigated here as well. The analysis of collisions reveals that particles interact prevalently via their flow fields rather than by direct contacts and a tendency towards particle clustering is demonstrated. The time evolution of the shape and size of particle tetrads is also examined.The simulations used in the aforementioned results were performed using a computational fluid dynamics simulation code that incorporates the Physalis method to apply particle boundary condition to the fluid fields. The implementation, which was developed by a previous graduate student, runs on a single graphics processing unit (GPU). In order to increase the possible domain size of the simulations, which is limited by the memory space on a single GPU, we present a new implementation of the simulation code that uses distributed memory to take advantage of supercomputers with hundreds of GPUs, allowing for much larger simulations. We document changes necessary to move to a distributed memory model for both the fluid and solid phases, and benchmark the code on up to one million resolved particles in a domain size of 1920^3 on 216 GPUs at the Maryland Advanced Research Computing Center. We also present favorable strong and weak scaling results, as well as validation of the implementation using two test cases.