The standard lattice Boltzmann method, which employs certain regular lattices coupled with discrete velocities as the computational grid, is limited in its flexibility to simulate flows in irregular geometries. To simulate large-scale complex flows, we present a cell-centered finite volume lattice Boltzmann method for incompressible flows on three-dimensional (3D) unstructured grids and its corresponding parallel algorithm. The advective fluxes are calculated by the low-diffusion Roe scheme, and the gradients of the particle distribution functions are computed with a least squares method. The presented scheme is validated by three benchmark flows: (a) a 3D Poiseuille flow, (b) cubic cavity flows with Reynolds numbers Re = 100 and 400, and (c) flows past a sphere with Re = 50, 100, 150, 200, and 250. Some parallel performance results are presented to show the scalability of the method, which reveal that the proposed parallel algorithm has considerable scalability and that the parallel efficiency is higher than 87% on 3840 processor cores. It can be seen that the presented parallel solver has significant potential for the accurate simulation of flows in complex 3D geometries.

中文翻译：

---

用于三维不可压缩流动模拟的可扩展并行非结构化有限体积格子 Boltzmann 方法

标准格子 Boltzmann 方法采用某些规则格子与离散速度耦合作为计算网格，在模拟不规则几何中的流动方面的灵活性受到限制。为了模拟大规模复杂流动，我们提出了一种用于三维 (3D) 非结构化网格上不可压缩流动的以单元为中心的有限体积格子 Boltzmann 方法及其相应的并行算法。平流通量采用低扩散 Roe 方案计算，粒子分布函数梯度采用最小二乘法计算。所提出的方案由三个基准流验证：（a）3D Poiseuille 流，（b）雷诺数Re  = 100 和 400 的立方腔流，以及（c）流过具有Re的球体 = 50, 100, 150, 200, and 250. 给出了一些并行性能结果来展示该方法的可扩展性，表明所提出的并行算法具有相当大的可扩展性，并且在 3840 个处理器内核上并行效率高于 87 % . 可以看出，所提出的并行求解器在精确模拟复杂 3D 几何中的流动方面具有巨大的潜力。