# Axial Flux Studio: reference-machine evidence register

Reviewed 2 November 2026 against the repository reference-machine notes, benchmark manifest, grading record and example source notes. Native 3-D figures are from the grid graded toward the iron, the default since 16 October 2026, unless a row says otherwise. These are selected documented development results, not fresh runs of the latest backend. Independent-input comparisons, fitted-model closures and derived references are different kinds of evidence, and none establishes a universal accuracy range.

Differences are (model − reference) / reference, rounded. Temperature differences are in kelvin. Records have different dates and conditions; they are not a controlled ranking between machines.

## 01: Parviainen, 5 kW TORUS-S

36 slots / 12 poles, one rotor between two slotted stators, full-pitch lap winding. Native 3-D, 16 October 2026; 300 rpm, magnets at 100 °C.

| Quantity | Model | Reference | Difference |
|---|---:|---:|---:|
| Phase fundamental RMS EMF | 214.1 V | 211 V measured | approximately +1.5% |
| Mean q-axis torque, 8.4 A | 171.3 N·m | 169 N·m inferred from measured EMF | approximately +1.4% |

Published geometry, turns and magnet data support an independent-input EMF comparison. The torque reference is derived from the measured EMF, not an independent shaft-torque measurement; the rated 159 N·m point has an unpublished current angle. The ungraded grid read 214.9 V and 171.9 N·m. An older section-model result of 242.8 V (+15%) used different modelling conditions and is not a convergence step.

Source: A. Parviainen, LUT doctoral thesis, 2005, Table 3.3. [Publication](https://lutpub.lut.fi/handle/10024/31185)

## 02: Sahin, 30 kW TORUS-S

24 slots / 4 poles, short-pitched lap windings, grain-oriented tape stators. Native 3-D; 1.3 mm as-built gap, both stators in series, magnets assumed at 25 °C.

| Quantity | Model | Reference | Difference |
|---|---:|---:|---:|
| Phase true RMS EMF per 1000 rpm (16 October) | 12.84 V | 12.79 V measured RMS | approximately +0.4% |
| Phase fundamental per 1000 rpm (2 October) | 12.76 V | the same 12.79 V | approximately −0.2% |
| Phase resistance, 26 °C (2 October) | 20.5 mΩ | 20.2 mΩ | approximately +1.4% |

The measurement's RMS convention (true or fundamental) is not stated, so both readings are shown against the same measured value; the fundamental row is not a separately measured fundamental. Bench magnet temperature is unpublished; ±10 °C corresponds to approximately ±1.2% EMF. The grain-oriented steel is modelled as isotropic, so this EMF result does not validate losses. The ungraded grid read 12.86 V.

Source: F. Sahin, TU Eindhoven doctoral thesis, 2001, Fig. 9.3 and Table 8.2. [Publication](https://pure.tue.nl/ws/files/2438215/200111643.pdf)

## 03: Ghent, 4 kW YASA

16 poles / 15 slots, whole-machine model with shod teeth.

| Quantity | Model | Reference | Difference |
|---|---:|---:|---:|
| Phase fundamental RMS EMF, 2000 rpm, native 3-D (16 October) | 126.9 V | 127.5 V | approximately −0.5% |
| Winding temperature, 100 W DC, 1000 rpm, lumped thermal network | 61.1 °C | 62.7 °C | −1.6 K |

The EMF is a closure check: the winding turns were fitted to EMF on the section model, so agreement tests the consistency of the reconstructed model rather than an independent prediction. The temperature comes from the thermal network with slot-face cooling and published support paths, with no fitting (reference: Energies 11(11):2983, Table 3); another measurement campaign reports 56.0 °C. Loaded magnet-temperature rises remain 60–85% low in documented cases.

Sources: Vansompel, Hemeida, Sergeant and collaborators; Mohamed et al., Energies 11(11):2983 and 11(4):774, 2018.

## 04: Abdelli / IFPEN, 18-slot / 12-pole TORUS-S

Ironless rotor, high-current reference. Slot dimensions (21 mm) and magnet grade (N42SH) are partly assumed.

### Loaded torque on the q axis

Native 3-D, graded axial grid, 8 October 2026; 1000 rpm, four rotor positions.

| Phase current, A peak | Model q-axis torque, N·m | Measured shaft torque + 1.8 N·m |
|---|---:|---:|
| 101.6 | 57.3 | 55.4 |
| 303.1 | 168.1 | 165.4 |
| 505.2 | 270.9 | 271.0 |
| 706.7 | 361.3 | 362.2 |

The experimental current angle is unpublished, so this curve uses only q-axis model values; the 100° benchmark below is not mixed into it. The four differences are approximately −0.2% to +3.4%, not a general accuracy range. The low-current samples span 53.9–60.0 N·m across rotor positions.

### Benchmark and losses

| Quantity | Model | Reference | Difference |
|---|---:|---:|---:|
| Electromagnetic torque, 1000 rpm, about 707 A peak, 100° electrical (16 October) | 370.5 N·m | 362.2 N·m | approximately +2.3% |
| No-load EMF fundamental, peak | 37.60 V | 38.40 V | approximately −2.1% |
| No-load loss, 4000 rpm (9 October) | 984 W | 1260 W, machined stator | approximately −21.9% |
| No-load loss, 6000 rpm | 2103 W | 2519 W, machined stator | approximately −16.5% |
| No-load loss, 8000 rpm | 3649 W | 4175 W, machined stator | approximately −12.6% |

The torque reference adds the 1.8 N·m no-load torque to the measured shaft torque. The loss points are separate no-load runs with the magnets segmented 15 times radially; figure readings are approximately ±2%. Measured losses are 14–28% above the model for the machined stator and 75–100% above it for stamped stators (the source's convention, with the model as denominator). Holder eddy currents, bearings and seals, and manufacturing damage are not modelled, so this is not an efficiency validation.

Source: Abdelli et al., EPJ Web of Conferences 330:01003, 2025 (CISTEM 2024), Figs. 11–12. [Publication](https://doi.org/10.1051/epjconf/202533001003)

## 05: Ghent, 26-pole / 24-slot YASA

48 V reference, 35 turns per tooth, eight coils in series per phase.

| Quantity | Model | Reference | Difference |
|---|---:|---:|---:|
| Phase resistance, 20 °C | 0.688 Ω | 0.742 Ω | approximately −7.3% |

An analytic winding-resistance check, not a newly rerun result. The reference is derived from measured coil resistance at 25 °C, temperature-corrected; it includes terminal wire, which the model omits. No measured EMF comparison is published, and magnet grade and arc are assumed.

Source: Van Damme, Vansompel and Crevecoeur, Machines 11(4):487, 2023. [Publication](https://doi.org/10.3390/machines11040487)

## Numerical checks are a separate question

Comparisons between the app's own solver formulations, and two-grid estimates, examine the numerical solution of one model. Measurements test the model together with its physical inputs. Agreement between formulations does not validate shared material or cooling assumptions, and a two-grid estimate is not a statistical confidence interval or an experimental accuracy guarantee.

Publication links identify the original work; they do not imply endorsement of Axial Flux Studio. These selected records are not a complete benchmark campaign.