Skip to the notebook

Can a thinner magnet make a propulsion motor lighter?

For an aircraft, a lighter propulsion unit can leave more mass available for payload or energy storage. Thinner magnets could help, but less magnetic flux can mean more current and heat at the same torque. Any saving has to survive the copper, cooling and retention needed around the magnet.

We use a to investigate how magnet thickness, rotor steel and winding choices interact. With the propeller and transmission fixed, the motor has to deliver the same shaft torque and speed.

We show the calculations, the checks that challenged them and the measurements that could change what we develop next.

In this notebook

Does thinning save mass at the same output?

Removing magnet material saves rotor mass. Does that saving survive the extra current, copper, cooling and retention the motor may need?

What must the magnet deliver when hot?

How much magnetic strength and resistance to demagnetization does the motor need at operating temperature? Can available magnets provide both?

Does a fault change the material requirement?

A short circuit changes the field opposing the magnets. Could it set a tighter requirement, and does changing the winding change that exposure?

Is a new material the right next step?

Could an available finished magnet or a different ring-and-winding combination meet the need? Which measurement would tell us where development should begin?

We modeled the reference in two dimensions, with assumed steel properties, windings and fault conditions.

Extra copper can outweigh the magnet’s mass saving.

Thinning the magnet from 1.5 to 1.0 mm takes 8.5 g out of the rotor. A thinner magnet also drives less flux across the air gap, so holding the same torque takes 11.5% more current. Copper loss grows with the square of current. Keeping it at the original level takes enough extra copper to outweigh the magnet material removed. Rotor and copper end up 4.0 g heavier.

Where the thinner magnet’s mass saving goes.

A 1.0 mm magnet against the 1.5 mm reference, with the original assumed ring.

Torque held across three direct solves
Winding comparison
← Lighter than referenceHeavier →
RotorMagnet and ring8.5 g lighter
Copper11.5% more current12.5 g heavier
Rotor + copperAt equal copper loss4.0 g heavier
4.0 g heavier. At 3.04 N m the thinner magnet needs 334 A-turns in each coil, 11.5% more. Holding copper loss takes 12.5 g more copper and raises slot fill from 29% to 36%.
Figure 01. Direct equal-torque solves; copper resizing is derived at equal loss. Equal loss does not establish equal temperature or thermal capacity. Cooling, retention and installation are excluded. Inspect the figure data

The ring behind the magnets carries flux between poles. In our assumed steel curves, the electrical steel saturates less readily than the stainless steel, so the motor needs less current at either magnet thickness. Keeping the original magnet then saves more mass than thinning it.

With the alternative steel, keeping the original magnet saves more mass.

← Lighter than referenceHeavier →
Thin magnet with original ring11.5% more current4.0 g heavier
Thin magnet with alternative steel5.7% more current2.5 g lighter
Original magnet with alternative steel5.1% less current5.1 g lighter
Original magnet with thicker original ring7.2% less current3.3 g heavier
Segmented, 1.5 mm with original ring6.4% less current; local field unresolved1.5 g lighter
Segmented, 1.5 mm without a ring4.2% more current; local field unresolved11.3 g lighter
Segmented, 2.0 mm without a ring10.5% less current; local field unresolved15.9 g lighter

On the alternative steel, the original magnet’s lower current allows either 5.1 g less copper at equal loss, or 19 W less copper loss with the original winding (20 °C resistance).

Figure 02. At 3.04 N m, 1.25 T and equal copper loss. Both steel laws are assumed. Hatched rows are segmented (Halbach) rotors with unresolved local fields and no retention. No row establishes installed mass. Inspect the figure data ; additional direct solves .
Why the winding belongs in the mass balance

Gemini 4 ArgonEarlier general review of component mass

Magnet and magnetic-circuit changes can enable lower surrounding steel, copper, or cooling mass even when current or copper heating binds.

GPT 6 AstraOriginal magnet with the alternative ring

If we retain the existing winding instead, that thicker-magnet alternative produces approximately 19 W less copper heat at the reference resistance, with no direct modeled mass saving. We cannot count both the full heat benefit and the copper removal.

The alternative ring lets us use less copper or generate less heat. The ledger counts the copper reduction at equal loss. Cooling and installation still have to fit within the remaining mass budget.

Direct equal-torque ledgerRing and winding comparison

Expand for Assumptions

The target is the model's 3.0445 N m mean torque, sampled over four rotor positions. It is a selected comparison load, not a measured continuous rating. The geometry derives from the , with 36 slots, 42 poles, a 10 mm stack and a 0.7 mm air gap.

At fixed turns, conductor length and copper temperature, equal copper loss requires copper mass to scale with squared current. It does not preserve hotspot temperature or transient thermal capacity. The steel laws are generic assumptions. Mechanical retention, cooling, losses, voltage headroom and production variation remain outside this mass comparison.

Current is peak ampere-turns per coil. With 25 turns per coil, 300 A-turns means 12 A peak phase current; it is not line RMS. Methods and inputs / Direct-solve report

A stronger magnet also needs to resist demagnetization.

A thinner magnet needs more remanence to produce the same torque per ampere. Remanence is the flux density remaining after the magnetizing field is removed. The magnet also has to resist the stator’s opposing field; beyond the knee of its curve, some magnetization is lost permanently. Both properties fall with temperature, and gaining one often costs some of the other. The figure compares three published specimens at 100 °C, including two made without heavy rare earths (HRE-free).

A quick picture of the magnetic knee

Strength and resistance to demagnetization are different properties.

MagnetizationHot kneeCoolerHotterOpposing field →
Schematic only. These drawn curves are not measurements. The supplier readings in Figure 03 define the knee at 90% of remanence.

Magnetic strength and the knee at 100 °C.

To keep a saving under ordinary load, a specimen has to sit right of the dashed line and above the solid one.

Inspect a historical specimen
Knee field at 100 °C (kA/m)6006507007501.211.221.231.241.251.261.271.28Remanence at 100 °C (T)45SH48SH52SH
1.221 T for zero rotor-and-copper savingOrdinary-load field from four joined direct solves
45SH (HRE-free) gives 1.227 T / 717 kA/m at 100 °C.Remanence is 6 mT above the conditional zero-saving boundary, inside the 7 mT reading resolution.The knee is 117 kA/m above ordinary-load field.Direct mass budget of 0.56 g before installation.

The dashed line compares with the original reference before installation, not with the original magnet on the alternative ring. The solid line covers ordinary load only. Neither is a qualified requirement.

Figure 03. Historical specimens at 100 °C, read from the GKN guide. Bars are reading resolution, not confidence intervals or batch variation. These grades neither isolate dysprosium’s effect nor define present supply. Inverse calculation ; specimen comparisons .

What can we buy before developing something new?

Figure 03 shows historical specimens, not today’s catalog. Two suppliers now list HRE-free grades. We have no hot finished-part data for either.

Bomatec N-52SH

The HRE-content sheet, p. 2 lists N-52SH as Dy/Tb/Ho/Gd-free.

We haven’t confirmed that its generic grade curves apply to this formulation.

CIBAS REN52SH Pure

The June 2026 catalog, p. 2 offers Pure grades excluding Dy/Tb/Ho/Gd through REN52SH.

CIBAS also lists a formulation containing Ho/Gd but no Dy/Tb. The catalog's general HRE-free definition is HRE content below 0.1%.

We haven’t qualified either route, and a missing public guarantee does not show a missing capability. Composition, hot finished-part performance, batch consistency, origin, orderability and procurement compliance each need their own check. Source update and limits.

Why we paused the proposed materials experiment

GPT 6 AstraThe proposed direction

The strongest hypothesis worth testing is a thin, HRE-free finished magnet that retains high hot remanence while gaining demagnetization resistance through grain-boundary processing.

GPT 6 AstraAfter checking existing supply

Existing HRE-free grades are credible enough that we should pause the proposed materials experiment. Our three-specimen comparison did not establish an industry capability gap. I moved too quickly from that comparison to grain-boundary development.

The next question became whether an available finished magnet, with a different ring and winding, could deliver the required performance. Hot finished-part data can distinguish a need for material development from a finishing or integration problem.

Supplier evidence and limitations

The winding matters during a fault, too.

A short circuit turns the spinning motor into a generator feeding its own winding. The current creates a field that opposes the magnets. Our first screen assumed twice the steady short-circuit current. Solving the field and winding circuit together gives a reverse-field estimate of 540–587 kA/m for the reference motor at three selected speeds.

The fault calculation changes our estimate of the opposing field.

Reference motor reverse field at the most loaded 1% of magnet (kA/m).

Earlier assumptionDoubled steady short-circuit current676 kA/m
450 rpmCoupled field + circuit solve540 kA/m
1,000 rpmCoupled field + circuit solve549 kA/m
2,000 rpmCoupled field + circuit solve587 kA/m

The dashed marker shows ordinary load alone at 532 kA/m.

Figure 04. Selected three-phase shorts at fixed speed, not a worst-case envelope. The assumption is a snapshot at one current; the coupled results follow each piece of magnet through time and retain its largest opposing field. Those are different statistics. Inspect the figure data .
Why peak current alone cannot set the magnet requirement

GPT 6 AstraAfter solving the fault through time

The five-rotor comparison is showing why a single current multiplier was inadequate. The field distribution and which magnet locations experience the peak matter as much as the peak current.

Claude Opus 5.5Earlier review of field exposure in a different motor

It is an exposure diagnostic, not a defeat condition; I had overstated it as one. Whether the magnet survives needs a measurement of the finished part's hot recoil and history after machining and coating.

We now retain each magnet location’s worst exposure before ranking the whole magnet. This corrects the field statistic, but calculated exposure is not measured irreversible damage.

Material-history and instantaneous field resultsNumerical checks and limits

Explore all 32 recorded faults and their time traces

Follow a recorded fault through time.

Radial with original ring, 1.5 mm, 2,000 rpmReference load and fault settingsCase ID sus15_2000_n8
Loading recorded trace…
587 kA/m following each piece of magnet through the event. The largest instantaneous screen is 566 kA/m. Peak phase current is 20.1 A.

The dashed line retains each piece of magnet’s largest field from the whole event, then ranks the pieces. The blue line ranks them again at each instant. The two differ because the most exposed piece can change with time.

Inspect all 32 recorded cases and assumptions

All cases assume isolated-neutral star windings, prescribed speed and linear magnet recoil. Connected trace segments join recorded steps.

Scroll horizontally to see all columns.

Case / rotorConditionsEvent field
(kA/m)
Peak phase
current (A)
sus15_450_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 300 A-turns before fault54012.0
refine_sus15_1000_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 300 A-turns before fault54914.1
pilot_sus15_1000_n4coarser mesh/time step0.894 Ω, 1.250 T, 20 °C winding, 300 A-turns before fault54914.0
sus15_1000_n16finest mesh/time step0.894 Ω, 1.250 T, 20 °C winding, 300 A-turns before fault54914.2
sus15_1000_n4_7cycles7 electrical periods; coarser mesh/time step0.894 Ω, 1.250 T, 20 °C winding, 300 A-turns before fault54914.0
sus15_2000_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 300 A-turns before fault58720.1
sus15_2000_endL0250.25 mH end inductance0.894 Ω, 1.250 T, 20 °C winding, 300 A-turns before fault56216.6
sus15_2000_hotwinding100 °C winding1.175 Ω, 1.250 T, 100 °C winding, 300 A-turns before fault56417.8
sus15_2000_noload_correctedno-load start0.894 Ω, 1.250 T, 20 °C winding, 0 A-turns before fault58318.6
sus15_2000_phase415° later fault start0.894 Ω, 1.250 T, 20 °C winding, 300 A-turns before fault58820.0
sus15_2000_phase830° later fault start0.894 Ω, 1.250 T, 20 °C winding, 300 A-turns before fault58719.3
sus10_1000_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 334 A-turns before fault62113.4
sus10_2000_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 334 A-turns before fault66717.7
steel15_1000_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 285 A-turns before fault53214.7
steel15_1000_n4coarser mesh/time step0.894 Ω, 1.250 T, 20 °C winding, 285 A-turns before fault53314.6
steel10_1000_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 317 A-turns before fault61412.9
steel10_2000_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 317 A-turns before fault67018.0
steel10_2000_n4coarser mesh/time step0.894 Ω, 1.250 T, 20 °C winding, 317 A-turns before fault66917.9
steel10_2000_n8_r2finer radial mesh0.894 Ω, 1.250 T, 20 °C winding, 317 A-turns before fault66917.9
steel10_2000_rewoundcopper resized for equal loss0.801 Ω, 1.250 T, 20 °C winding, 317 A-turns before fault68118.7
steel10_2000_n4_7cycles7 electrical periods; coarser mesh/time step0.894 Ω, 1.250 T, 20 °C winding, 317 A-turns before fault66917.9
steel10_45SH_hot_2000_rewound1.227 T remanence; copper resized for equal loss; 100 °C winding1.018 Ω, 1.227 T, 100 °C winding, 322 A-turns before fault64616.8
sus_thickring_1000_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 278 A-turns before fault53214.9
sus_thickring_1000_n4coarser mesh/time step0.894 Ω, 1.250 T, 20 °C winding, 278 A-turns before fault53114.7
halbach_sus15_1000_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 281 A-turns before fault114814.4
halbach_sus15_1000_n4coarser mesh/time step0.894 Ω, 1.250 T, 20 °C winding, 281 A-turns before fault126814.3
halbach20_1000_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 269 A-turns before fault100716.2
halbach20_2000_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 269 A-turns before fault101422.9
halbach20_2000_n4coarser mesh/time step0.894 Ω, 1.250 T, 20 °C winding, 269 A-turns before fault108222.7
halbach20_2000_rewoundcopper resized for equal loss1.115 Ω, 1.250 T, 20 °C winding, 269 A-turns before fault101320.8
halbach15_1000_n8Reference load and fault settings0.894 Ω, 1.250 T, 20 °C winding, 313 A-turns before fault100713.6
halbach15_1000_n4coarser mesh/time step0.894 Ω, 1.250 T, 20 °C winding, 313 A-turns before fault108213.4
All 32 completed cases, ending at their last recorded step. The field screen is not measured damage or a predicted 1% flux loss. The failed no-load attempt and corrected rerun remain in the scientific record. Case inputs and results .

The added copper also lowers winding resistance, which can raise fault current. For the thin magnet on the alternative ring at 2,000 rpm, the equal-loss rewind raises the reverse-field estimate from 670 to 681 kA/m.

The segmented rotors’ local fields kept changing under mesh refinement, so we reject those thresholds. An earlier blind test of torque and voltage on two other motors failed, and a separate fixture comparison was inconclusive. We still haven’t validated the motor's internal fields.

Inspect the refinement and validation checks

Refining the mesh and time step barely moved the fault current. In the segmented rotors it moved the local field 7–10%, beyond the frozen 3% limit, so we can’t use those values as exact field thresholds.

Current can converge while the local field still changes.

Change on refinement (%). The local field is the quantity the magnet has to survive.

Radial magnetsWithin criterion
Current 1.1%Local field <0.1%
Segmented, no ringFailed criterion
Current 1.3%Local field 7.5%
Segmented, with ringFailed criterion
Current 1.1%Local field 10.4%
CurrentLocal field3% frozen limit on a 0–12% scale
All nine refinement comparisons, including failures

Scroll horizontally to see all columns.

Coarser → finer caseCurrent changeField changeCriterion
pilot_sus15_1000_n4
→ refine_sus15_1000_n8
1.10%0.01%Within 3%
refine_sus15_1000_n8
→ sus15_1000_n16
0.43%0.03%Within 3%
sus_thickring_1000_n4
→ sus_thickring_1000_n8
0.97%0.10%Within 3%
steel15_1000_n4
→ steel15_1000_n8
1.05%0.05%Within 3%
halbach15_1000_n4
→ halbach15_1000_n8
1.28%7.50%Failed 3%
halbach_sus15_1000_n4
→ halbach_sus15_1000_n8
1.10%10.44%Failed 3%
steel10_2000_n4
→ steel10_2000_n8
0.42%0.22%Within 3%
halbach20_2000_n4
→ halbach20_2000_n8
0.86%6.76%Failed 3%
steel10_2000_n8
→ steel10_2000_n8_r2
0.07%0.09%Within 3%
Figure 05. Matched mesh and time-step refinement, with the 3% criterion fixed before comparison. This is numerical agreement within the model, not physical validation. The segmented local-field thresholds are rejected. All refinement records .

The fixture’s point estimates passed. The band was too wide.

Drop in gap field under opposing current (mT). Shading is an assumed sensitivity band, not a confidence interval. Symbols are separated vertically for clarity.

01002003004002,000 A-turns4,000 A-turnsDrop in gap field (mT)
PredictedMeasuredSensitivity band
Point errors3.9% / 6.2%

Both meet their central-point tolerances.

Band at 4,000 A-turns170 mT

Fails the 60 mT width limit.

This check was inconclusive. It does not replace the failed blind torque/voltage test on two other motors or validate the motor’s internal reverse field. Now that its results are released, it cannot serve as a new blind test.

Inspect the predictions, readings and sensitivity bands

Scroll horizontally to see all columns.

Opposing currentPredicted dropMeasured dropSensitivity bandWidth criterion
2000 A-turns138.7 mT144.4 mT97.6–182.5 mTNot tested
4000 A-turns277.5 mT295.8 mT195.3–365.1 mTFailed
Figure 06. A simple magnetic-circuit model predicts reversible field in a separate fixture gap, not the motor’s internal reverse field. Measured values from Sjokvist and Eriksson (2014). Their plot-reading bars are ±10 mT; the two differences share a zero-current reading and have correlated errors. Released validation score .

An independent validation agent selected the published fixture, withheld its outcomes from the builder and froze the criteria before predictions. The fixture checks reversible gap field; it does not validate hot NdFeB, irreversible damage or the motor's internal fields.

A simplified axial calculation gives about 6% less received flux in one 10 mm geometry; it is not a correction factor for this motor. A separate conducting-ring calculation passed numerical controls but could not resolve its field effect under its frozen criterion. Its outcome is also inconclusive.

Independent scientific review / Fault refinement review / Full scientific closure

Toward a lighter propulsion unit

The investigation made the material requirement more specific. A useful magnet has to stay strong at operating temperature, resist the opposing fields it meets under load and during a fault, and not cost more in copper, cooling and retention than it saves.

Catalog grades may meet the magnetic requirement on paper. The open question is the finished part. If hot magnetic performance limits the design, comparing a finished magnet arc with a parent sample from the same batch at operating temperature can separate a material limitation from changes introduced during processing. If current or heat rejection limits it, measuring the assembled motor at matched output may tell us more first.

That evidence decides whether we develop the material, the finishing process or the way the component is integrated. If you’re interested in pursuing these open questions with us, reach out. We’d love to chat.

Data, methods and checks.

The figures use the original calculation tables. Both scientific snapshots are preserved, including failed runs and the criteria used to judge them.

Methods are plain-text Markdown; their relative paths refer to the original scientific record.

Sources, rights and reuse

INCLUDED as derived geometry and outputs. Digital files permitted with attribution; source files not included.

Koo and Shin 2026 prototype measurements

INCLUDED. Table measurements permitted with attribution; extracted numeric rows included.

GKN permanent magnet performance guide

SELECTED NUMERICAL OBSERVATIONS INCLUDED; full curve dataset withheld. PDF plot artwork and full digitized curves not redistributed; broader reuse unresolved.

Dynamical public-reference code and calculations

INCLUDED. Corrected solver and retained numeric outputs included.

Sjokvist and Eriksson2014 public author postprint

SELECTED NUMERICAL OBSERVATIONS INCLUDED. PDF, page images and full vector plot are not redistributed.

Our code, calculations and presentation are identified as Dynamical's work, shared for inspection; this page does not grant a blanket license over the package. Third-party measurements and documents retain their own terms. Supplier PDFs, full curves and source artwork are not redistributed. Complete source-rights register.

Failed checks and limits on our claims
  • A thinner magnet necessarily makes the installed unit lighter.
  • The doubled-current case is a predicted or guaranteed fault transient.
  • Stable torque/current alone validates local magnet field.
  • A 1% field-exposure screen predicts 1% irreversible flux loss.
  • The simplified axial ratio is a correction factor for the native motor.
  • The selected typical curves are finished-part, batch, domestic-origin or procurement guarantees.
  • A failed field refinement proves the physical construction infeasible.
  • The new fixture comparison validates the native motor or repairs the original failed two-motor evaluation.
  • Equal copper heat establishes equal hotspot temperature after changing conductor mass.

The archive preserves the original failed two-prototype evaluation, the failed no-load attempt and its separate repaired rerun, and the failed segmented-field refinements.