Workflow
Motion, contact & flow
Nothing is prescribed here. You place bodies, give them a rotation and material coefficients, and the trajectories, contact forces, orbit rates and passing frequencies come out of time integration. A bearing's cage speed is an output, not a formula the solver was handed.
1. Setting up a motion scene
There is nothing to set up: every part in the scene is already a body, and what it does follows from what you attached to it. You do not sort parts into classes.
- A part a support holds stays where it is: it pushes back what hits it, and turns the blows it takes into sound through its own modes.
- A part a rotation drives turns: its contact points move at that speed and drag what touches them through friction.
- A part with neither falls: it collides through its own triangle mesh, with mass and inertia measured off its geometry. If that mesh measures out as a sphere, the solver says so and uses the exact analytic Hertz sphere contact for it.
- A part with Use in simulation switched off takes no part in any solve.
So you make a part stand still by supporting it, not by marking it "static". Which part gets struck, and which one makes sound, is not chosen either: it is what the run produces.
Two coefficients on the part itself set its impact sound: Restitution e (energy kept in a collision) and Friction μ (rolling and sliding noise). Defaults are 0.6 and 0.4.
Rotation is a scene object, not a solve parameter. Scene ▸ Add Rotation asks for Speed in rpm and an Axis, and every solve reads it off the scene. Gravity is scene state too (Scene ▸ Gravity), and so are gas springs (Scene ▸ Gas Spring) and sealed cavities (Scene ▸ Sealed Cavity), which enter the solve through the gas law rather than through a stiffness you type. Where two parts meet on cylindrical surfaces, Fit contact surface measures those surfaces from the mesh: pick Surfaces ▸ Inner surface (hole) or its outer counterpart.
2. Running a motion simulation
Either check How do the parts move and collide? in Simulation ▸ Solve ▸ ▶ SOLVE, or run it on its own from Individual solvers ▸ Motion Simulation… for the full detail dialog. (Under ▶ SOLVE, tick Ask me for each step's settings (advanced) to get the same dialog; left unticked, each step runs with defaults measured from the scene.)
The dialog's choices, and what each costs:
- Recording duration: seconds of recorded motion, from t = 0 with the bodies where they stand. Default 4 s.
- Motion solver: Fast or Multi Body Dynamics. The fast path is the quick look; Multi Body Dynamics integrates the contact forces in time. Aerodynamics runs on the Multi Body Dynamics path only. The two settings below appear only when it is picked.
- Start state: At the rolling state (recommended) seats a driven contact train in its steady rolling state at t = 0, so a short run does not spend itself on spin-up. Use At rest (spins up slowly) when the transient is the subject.
- Contact stiffness: Full Hertz (small step, accurate) uses the real contact stiffness and a time step around 1e-7 s, which is 50 million sub-steps for a 5 s run. Softened (large step, fast) trades that for a bigger step. Softening changes the contact frequency content, so compare like with like.
- The falling parts' sound: Include it (impact sound + ring-down) makes each part push air when it strikes and ring with its own modes if it has them. Only the held parts' sound leaves that out.
Press Simulate. The run reports orbit rate, contact-pass frequency and orbit radius, all outputs. The record lands in the tree with its impact and track counts, and Play motion replays the bodies together with the sound from the same timeline.
3. Turning motion into sound, later
Motion and sound are separate steps. The stored motion record can be radiated again without re-running the physics.
- Solve the modes (Solve Modes); the acoustic step needs them, the motion step does not. If a part the run touched has no modes when you radiate, its modes are solved right then (provided it has a volume mesh) and you are told which ones.
- Either press Re-radiate on the motion record's row in the tree, or open Choose acoustic method… and pick Motion → sound: contact events + FEM modal, choosing the Motion record to radiate (or New motion simulation… to make one now).
- Choose the Radiating mode set and the Radiation method: the same Rayleigh / shadowing / BEM ladder described in Sound radiation & rooms.
Impact sound is built from Hertz contact mechanics, and thousands of impacts are summed in one solve. Squeal and large plastic deformation are out of scope.
4. The coupled flow solver
Adding air to a rotating scene turns the run into a coupled lattice-Boltzmann solve with a sharp interface: blade loads, shaft torque and flow rate are computed, with no drag coefficient entered anywhere. The price is wall-clock time, quoted before you commit.
- Give the scene a rotation (Scene ▸ Add Rotation). Tip speed is measured from it and sets the grid and the time step. Without a drive there is nothing to size the run against.
- Add the domain: Scene ▸ Air Volume, which opens Air volume (flow domain). Type the box only: Edge X, Edge Y, Edge Z and the Centre x, Centre y, Centre z. Cell size is a preset rather than a number: Resolution ▸ Automatic (from the rotor radius), Development: 6 mm or Full: 3 mm. The time step is not offered at all: sound travels 1/√3 cells per step in the lattice, and a longer step would put the microphone output at the wrong speed of sound. The sponge layer that absorbs reflections is added outside your box, so the volume you typed is not clipped.
- Read the cost estimate, which appears as soon as the volume is attached: what one solved second costs, in millions of cells and the grid shape, Δx in millimetres, memory in GB, and hours. Tip speed and tip Mach number come with it. If the scene cannot carry a flow solve at all (no drive, Mach out of range, no usable GPU) you get that reason instead of a number.
- Run it. From ▶ SOLVE, check How does it move the air? (thrust, flow rate, flow noise); both boxes ticked still run once. From the motion dialog, set Aerodynamics (flow solve) to On, development resolution (~6 mm) or On, full resolution (~3 mm) instead of Closed: mechanical only. A cell size you typed on the air volume wins over the preset, and the run says which grid it used.
GPU acceleration. Settings ▸ Simulation Settings ▸ General ▸ GPU acceleration selects hardware, not physics, and is machine-local rather than stored in the project. Small jobs stay on the CPU on purpose: below roughly ten million receiver × element × frequency products the GPU is up to ten times slower. Above it, 20–40×. If GPU memory runs out the solve falls back to the CPU rather than failing.
Absolute thrust is not validated. The boundary layer is not resolved at these cell sizes, and thin-section lift is read low. On a NACA 0012 section against thin-airfoil theory the lift coefficient reads −43 % at 20 cells per chord and gets worse with refinement (−52.5 % at 80). On the Caradonna–Tung rotor the same chain reaches about 94 % of the literature reference. The shipped aero example is therefore a comparison: two configurations, same grid, same solver, same rpm, same bias. It claims the difference, not the absolute number.
Mesh independence is shown for the sphere only. A Richardson study exists for sphere drag. The fan fleet has no frequency–mesh convergence study yet, and one rotor grid refinement moved C_T by tens of percent without converging monotonically. Refine your own case before quoting an absolute.
The far field is a compact Curle dipole. Loading noise propagates; turbulence's quadrupole contribution does not travel to the far microphone at all. Broadband sources are under-represented.
Runtime and memory are real. A rotor case at 9.5 mm cells is 35.6 M cells for 0.72 s of physics: roughly ten minutes on an idle card, 55 minutes on the same card with other work on it. There is no checkpointing, so a run that diverges at 80 % loses everything. Read the cost card before you start, and start with development resolution.
5. Reading flow results
Flow results are viewed in the 3D scene through the Layers menu, split into Structure and Flow. The flow layers are independent checkboxes:
- Air (particles): tracer particles carried by the field, with trail length derived from the run. The most direct way to see recirculation and a wake settling.
- Slice (plane): a cross-section coloured by the field, for reading magnitudes.
- Streamlines (mean): the time-averaged field. Instantaneous streamlines in an unsteady wake mislead.
- Vortex (Q): a Q-criterion iso-surface, where rotation dominates strain. Shows blade tip vortices and how far downstream they survive.
- Volumetric: the whole field rendered through, for extent rather than detail.
Colour source switches every flow layer between Pressure and Velocity. Playback uses the transport under the view: Play / Pause, One frame back, One frame forward, a Timeline scrubber and a speed slider, with the frame counter beside it. The colour bar's range is editable, so type the top and bottom numbers to hold a scale across two runs. The colormap chooser offers Sequential (viridis) and Diverging (two-ended); pressure is signed, so it belongs on the diverging map.
A coupled run also produces sound: ordinary virtual microphone signals in the tree, playable and exportable through Listen / download ▸ ▶ Play, WAV or CSV. The caveat above applies, and what the compact dipole carries well is the tonal, loading-noise part.
Check a cage speed against a formula the solver never saw. On a bearing rig, compute the fundamental train frequency and the ball-pass frequency from the geometry by hand, then compare them with the orbit rate and pass frequency the run reports. In the shipped benchmark the emergent values match the kinematic formulas to +0.01 % (11.894 vs 11.895 Hz for the cage; 107.046 vs 107.053 Hz for ball-pass), with no force and no frequency entered anywhere. If your own rig disagrees badly, check the contact surfaces and the start state first.
Halve the time step and see whether the answer moves. The measured acceptance criterion on the bearing rig is that halving the step (1e-6 → 5e-7 s) moves the emergent cage rate by less than 1.5 %. Apply the same test to an impact case: run it, halve the step, and compare the impact count and the microphone SPL. If either moves appreciably, the step was doing the talking, or Softened (large step, fast) contact was.
See also: Sound radiation & rooms for the radiation ladder, Settings, units & licence for GPU and the SI rule.