CC0 · public domain AXF-1 · rev A/ outer-diameter sizing sweep/ self-consistent from §REL

AXF-1 — how big should it be?

The sweep holds the electromagnetics fixed (Bg=0.40 T, current density, fill, axial stack) and scales only the outer diameter. At constant airgap shear stress the four quantities that decide the actuator each scale as a clean power of D — which on the log plot below are four straight lines fanning out from the 110 mm baseline. There is no electromagnetic optimum: the binding constraint is elsewhere.

CURVE

Everything scales as a power of diameter

log–log · normalized to 110 mm = 1×
0.25× 0.5× 8090 100110 120130 140150 160 outer diameter Dₒ (mm) 110 mm = 1× Kₘ=1.3 → 122 mm rotor inertia ∝ D⁴ torque ∝ D³ Kₘ & mass ∝ D² torque density ∝ D¹
Fig.2 — Slope = scaling exponent. The payoff (torque, ∝D³) rises fast, but the dynamic cost (rotor inertia, ∝D⁴) rises faster — it is the steepest line on the chart. Km and mass rise together (∝D²): the motor constant is bought in proportion to added mass. All curves ride on Bg=0.40 T; a lower measured Br shifts them down together without changing the slopes.
DATA

Absolute values at key diameters

joint torque after 8:1 × 0.85
ODmotor mass T air (6)T nom (9)T peak (18) KₘTD peakinertia
100 mm0.70 kg22.233.366.60.8713.92.4e-4
110 mm0.85 kg29.644.388.71.0615.33.5e-4
120 mm1.01 kg38.457.6115.11.2616.75.0e-4
130 mm1.19 kg48.873.2146.41.4818.16.8e-4
140 mm1.38 kg60.991.4182.91.7119.59.2e-4
Units — torque N·m · Kₘ N·m/√W · TD (motor torque density, peak) N·m/kg · inertia (rotor active) kg·m². Cooling column = continuous current density in A/mm². Nominal (9) is the datasheet operating point (σ ≈ 11.6 kPa, P_cu ≈ 38 W) and needs light forced convection; air (6) is passive-continuous.
READ

What the curve actually decides

▲ bigger is strictly better — electromagnetically

Torque ∝ D³, Km ∝ D², torque density ∝ D¹ — all monotonic. There is no peak to find; the physics never says "stop." Doubling the payoff costs only a squared growth in mass.

It also wins on your ledger: Km rising means less copper loss per newton-metre, so a larger machine is more joules-efficient per unit torque. Energy-optimal points the same way as torque-optimal — up.

▼ what stops you is inertia and packaging

Rotor inertia ∝ D⁴ — the steepest line. Reflected through the gearbox (×G² = 64), it dominates how fast the joint can accelerate. At 110 mm the reflected rotor inertia is ≈ 0.022 kg·m²; at 140 mm it is ~2.6× that. Past some diameter you add inertia faster than torque and the limb goes sluggish.

And the actuator sits at the joint, so its mass is distal load — every gram penalizes the limb it drives and the proximal joints that must swing it. The ceiling is a packaging + dynamics call, not a magnetics one.

The decision, as a curve

Hip / knee (needs torque, tolerates mass): 120–130 mm is the sweet spot — 130 mm buys 73 N·m nominal / 146 N·m peak and clears the Km ≥ 1.3 datasheet target (crossed at 122 mm) at 1.19 kg, if it packages into the thigh. Ankle / wrist / elbow (needs low distal inertia): stay at 100–110 mm and accept less torque — the D⁴ inertia term matters more distally than the D³ torque gain.

So AXF-1 is not one motor — it is one design scaled per joint: same poles, same coils, same magnets, same process, only Do changes. That is the printability dividend — you re-slice, you don't re-engineer.

Correction the sweep forced. The rev-A datasheet targeted ≤ 0.60 kg motor mass at 110 mm. Self-consistent, the honest number is ≈ 0.85 kg (Cu 183 g + Fe 240 g + magnet 186 g + 40% structure). Motor-level torque density is therefore ~0.35× sintered, not the 0.55× first quoted — the sovereignty tax is larger than stated. The joint-level and Km stories are unchanged. Fold 0.85 kg into rev B before the RFQ goes out.