Inner and outer polarity
Outer ring · #01Outer poles face inward; inner poles face outward. Click to cycle N / S / empty. Press 1 or 2 to switch rings.
Tune both magnetic rings · Estimate drag and smoothness from hands-on measurements · Hear the linked sample
Outer poles face inward; inner poles face outward. Click to cycle N / S / empty. Press 1 or 2 to switch rings.
By default, group B/C drag scores are multiplied by 2.8 ÷ 2.5 = 1.12; smoothness scores stay unchanged. This aligns the reported nine-magnet feel with the factory rating. Disable it to use the original PDF scores. The PDF factory overall-feel rating of 9 is not a smoothness measurement.
New measurements are saved in this browser and included in exported configurations.
| Configuration | Original PDF drag | Calibrated drag | Smoothness | Evidence | Action |
|---|
The updated GLB is the complete assembly. The program retains its triangles and assembly transforms, shares repeated geometry, and quantizes each coordinate axis to under 0.001 mm error. Colors are adjusted for clarity; the original CAD shapes remain.
The radial-dipole approximation uses 1.5 mm of softening; near-field accuracy for finite magnets is limited. Features include magnet counts, net polarity, neighboring correlations, spatial harmonics, and conservative torque variation. Three kernel widths show model disagreement. Nearby observations correct the prior; a measured layout recovers its calibrated score. The prior coefficients are modeling assumptions, not uniquely inferred material constants.
Conservative magnetic forces can create attraction or cogging but do no net work over a turn. Sustained drag may include eddy currents, hysteresis, contact friction, and collision losses. The PDF does not separate these terms, so scores are not physical torque measurements.
Residuals are extrapolated using spatial-layout similarity, not count alone. An empty ring has zero magnetic contribution (bearing friction can remain). Reversing both rings or shifting the slot origin leaves the full-turn estimate unchanged.
Displayed ranges combine subjective rating uncertainty, configuration distance, and model disagreement; they are not statistical confidence intervals. The 2N–2S pattern and most 15–27 magnet layouts lack measurements, so their ranges are wider. These records do not establish that odd counts always win or that a specific count is optimal.
The attached recording supplies the same segment around 22–27.5 s. Short clicks are triggered by relative ring phase, local polarity interaction, and estimated loss. Pole passes can overlap, so the total encounter rate is not the perceived pitch.
Force at a known line radius, measured speed, and coast-down curves would help convert subjective scores to torque. Exact slots, pole directions, and recordings are needed to test new-layout predictions.
References: COMSOL Magnetic Brake (eddy-current losses and motion); SciPy RBFInterpolator (radial-basis interpolation; implemented locally in JavaScript).