Drone motors designed
backwards.
Most motors are designed the same way: wind a prototype, test it, adjust, repeat until it works well enough. FRA142 works backwards. Geometry comes from target performance, not the other way around. The simulation runs thousands of configurations. Only then does anything get manufactured.
The MEC solver evaluates a candidate motor in under 1 ms. That speed is what makes optimization practical: you can run 12,000 candidates in a loop without waiting overnight. Calibrated against 130 commercial motors from 28 brands.
Works like natural selection, applied to motor geometry. Start with 80 random candidates. Score each on efficiency and mass. The best survive, recombine their traits, mutate slightly. After 150 generations the population converges to a Pareto frontier: designs where you can't improve one metric without hurting the other. The Gabizos-6LR was picked from that frontier.
Every part is generated in CadQuery from parameters, not drawn by hand. Change a blade angle and the STEP files update automatically. The motor base went through topology optimization (SIMP + PyTorch) and lost 70% of its mass in the process.
Each motor in the FRA142 lineup started as one of 80 random geometries. The optimizer treats motor design like breeding: score each candidate on efficiency and mass, keep the fittest, let them recombine and mutate, repeat. This is NSGA-II, a multi-objective genetic algorithm.
10 design variables are explored simultaneously: stator dimensions (OD, stack length, bore, tooth width, yoke, tooth height), rotor parameters (air gap, magnet thickness, bell thickness), and winding (turns per coil). Every combination that violates a physical constraint (saturation, thermal, fill factor) is penalized but not killed, so the search can explore boundary regions.
After 150 generations, the population converges to a Pareto frontier of ~50 motors. On that frontier, no motor beats another on both efficiency and mass at the same time. Every point is a real, physically valid design. The production motor is selected from this frontier and then adjusted for what a machine shop can actually build.
The MEC solver was validated against 130 commercial motors from 28 brands, not just a handful of in-house designs. On the FPV segment it was actually built for (2207–2807 stators), Kv prediction error is 2–5%.
Across the full 130-motor dataset (including DJI and other segments outside the target), 80.8% of motors land within ±10%. All 192 unit tests pass. The pipeline is reproducible.
| Format | 23 × 10 mm (2310) |
| Configuration | 12N14P outrunner |
| Weight | 36 g |
| Kv | 1287 RPM/V |
| Voltage | 5S – 6S LiPo |
| Prop | 6" long range |
| Magnets | N52 × 14 arc |
| Stator | M19 silicon steel |
| Winding | 12 turns · Ø 0.50 mm |
| Peak power | 550 W |
| Point | RPM | η | P_out |
|---|---|---|---|
| cruise 4S | 14,800 | 83.5% | 38.7 W |
| cruise 5S | 18,500 | 84.1% | 58.1 W |
| cruise 6S | 22,200 | 84.5% | 81.4 W |
| sprint | 28,000 | 78.8% | 234.6 W |
| full power | 35,000 | 72.3% | 550 W |
| Constraint | Limit | Margin | |
|---|---|---|---|
| bell saturation | < 1.5 T | +39% | [PASS] |
| tooth saturation | < 1.8 T | +20% | [PASS] |
| airgap | ≥ 0.15 mm | +33% | [PASS] |
| fill factor | ≤ 50% | +30% | [PASS] |
| magnet temp | < 80°C | +43°C | [PASS] |
| winding temp (60s) | < 120°C | +22°C | [PASS] |
> interested in the Gabizos-6LR?_
Five prototypes are going into production.
If you want to bench-test one or just follow the build — reach out.