Oil-Bath Lubrication of Helical Gears

Where the Oil Actually Goes

In a lubricated gearbox the oil bath is not a boundary condition — it is a moving, splashing free surface that decides which tooth flanks get replenished, how quickly, and at what cost in power. That is awkward to measure: the region of interest is a thin film inside a mesh that is continuously opening and closing, dragging oil in and throwing it back out. Simulation is the practical way in.

Coupling SPH to a Finite-Element Gear Mesh

The gear side of the problem is handled by MUTANT, a gear contact code whose development the author began during his PhD at KU Leuven's LMSD division and which has since been further developed by PhD researchers there, largely under his supervision. It resolves the finite-element deformation and contact of the meshing helical gear pair. For this study its FE solver was coupled to a Smoothed Particle Hydrodynamics (SPH) solver carrying the oil.

SPH is a mesh-free, Lagrangian method. The oil bath is discretised into particles, each carrying physical properties — mass, momentum, density and pressure — and the 3D Navier-Stokes equations are approximated by evaluating interactions between neighbouring particles: every field quantity at a particle is a kernel-weighted sum over the particles inside its support radius, with the pressure and viscous terms following from the gradients of that same kernel. Because there is no mesh, there is nothing to distort or remesh when the fluid tears into sheets and droplets, and the free surface needs no explicit tracking — it is simply wherever the particles are. Fluid motion is therefore represented directly by particle trajectories, which were imported into Blender for the render.

Trust in the approach was built rather than assumed, through sanity checks, convergence studies and experimental validation.

Oil Supply, Replenishment and Losses

Oil-bath simulations of this kind are standard practice in lubricated drivetrain analysis, for two reasons. They determine the oil supply and replenishment paths — how much oil reaches each flank and how fast it returns once squeezed out — which is precisely the inlet condition that finer-scale Elasto-Hydrodynamic Lubrication (EHL) simulations require. And they make the lubricant-related power losses predictable: churning, windage and squeezing, none of which a gear-only model can produce.

A Note on Visual Plausibility

The same simulation doubled as a small experiment. A video of the corresponding non-lubricated gear simulation was given to a generative video model with a prompt to add physically accurate oil splashing; the animation shows both, the physics-based result above and the generated clip below. Roughly half the people shown the pair picked the generated one as the simulation. The interesting part is not that a generative model gets conservation laws or churning losses wrong — with enough domain-specific training data and physics-informed loss functions, machine-learning surrogates will do considerably better. It is the visual plausibility trap: trusting a simulation because its visualisation looks convincing and advanced. That trap is as old as computational mechanics itself; generative AI has only made it cheaper to fall into.

Significance

An oil-bath simulation closes a gap in the drivetrain modelling chain. Macro-scale gear contact analysis says where and how hard the teeth press together; EHL says what the film does under that pressure; the fluid simulation in between says whether there is any oil there to form a film in the first place. Resolving that middle step is what turns a lubrication assumption into a predicted loss — the difference between setting an oil level by convention and setting it by analysis.