Benchmarking of Computational Fluid Methodologies in Resolving Shear-Driven Flow Fields

Abstract

Despite recent interests in complex fluid–structure interaction (FSI) problems, little work has been conducted to establish baseline multidisciplinary FSI modeling capabilities for research and commercial activities across computational platforms. The current work investigates the fluid modules of contemporary FSI methodologies by solving a purely fluid problem at low Reynolds numbers to improve understanding of the fluid dynamic capabilities of each solver. By incorporating both monolithic and partitioned solvers, a holistic comparison of computational accuracy and time-expense is presented between lattice-Boltzmann methods (LBM), coupled Lagrangian–Eulerian (CLE), and smoothed particle hydrodynamics (SPH). These explicit methodologies are assessed using the classical square lid-driven cavity for low Reynolds numbers (100–3200) and are validated against an implicit Navier–Stokes solution in addition to established literature. From an investigation of numerical error associated with grid resolution, the Navier–Stokes solution, LBM, and CLE were all relatively mesh independent. However, SPH displayed a significant dependence on grid resolution and required the greatest computational expense. Throughout the range of Reynolds numbers investigated, both LBM and CLE closely matched the Navier–Stokes solution and literature, with the average velocity profile error along the generated cavity centerlines at 1% and 4%, respectively, at Re = 3200. SPH did not provide accurate results whereby the average error for the centerline velocity profiles was 31% for Re = 3200, and the methodology was unable to represent vorticity in the cavity corners. Results indicate that while both LBM and CLE show promise for modeling complex fluid flows, commercial implementations of SPH demand further development.