Multi-Phase Gearbox Modelling Using GPU-Accelerated Smoothed Particle Hydrodynamics Method

Abstract

Developments in automotive design such as electrification of engines and a growing need to improve driveline efficiency requires adaption of old techniques. The ability to make fast and accurate Computational Fluid Dynamics (CFD) assessment is of high importance to the development of novel powertrains. Consequently, innovative numerical techniques and continuous improvements to existing CFD codes is relevant to ensure reliability. This work extends the capabilities of a Smoothed Particle Hydrodynamics (SPH) code to include multiphase modeling, studied using a gearbox model. A vast majority of CFD codes use grid-based approaches following the Eulerian spatial discretization, which is quite established in engineering applications. Lagrangian based approaches where the moving fluid particles are discretized over time and space present a promising alternative. One of the most common methods of this kind is the Smoothed Particle Hydrodynamics (SPH) method, a fully Lagrangian, particle-based approach for fluid-flow simulations. The main advantage is the absence of numerical grid for computations, which eliminates complexities of interface handling. Nowadays, the SPH approach is more commonly used for hydro-engineering applications involving free-surface flows. New techniques to perform numerical simulations on Graphics Processing Units (GPU) virtually eliminates some of the disadvantages of the method. In this work, we present our multi-GPU solution designed for both GPU-equipped desktops and multi-GPU supercomputers. Fluid dynamic simulations on a single gearbox model is used to validate the multiphase model, by comparing the results with earlier simulations that use a single-phase model omitting air-lubricant interface in the gearbox. The base case in the study is a single bevel gear placed inside a cuboid case with a lubricant depth equivalent to 25% gear diameter. Simulations are performed at various rotational speeds, and corresponding lubricant distribution and churning losses are obtained. The current study targets a comparison of the single-phase and multiphase models in approximating the lubricant distribution and churning loss values at nominal rotational speeds. This serves to standardize the numerical procedure, which will help in improving the accuracy of churning loss calculations through validations against experimental results in the future.