phasesweep

PMSM simulation for stator/rotor design exploration. Built on motulator and NGSolve.

What it does

• Drive simulation — SPMSM and IPMSM comparison using motulator (current-vector control, mechanical load)
• 2D FEM field solver — NGSolve magnetostatic solver with smooth-bore and slotted stator geometries, linear or nonlinear iron (Picard iteration)
• Analytical validation — Zhu & Howe (1993) closed-form air-gap field model for cross-checking FEM results
• Slotted stator analysis — OCC boolean geometry for exact arc-bounded slots, armature reaction decomposition, slot-count sweeps
• Parameter sweeps — explore the flux-linkage (psi_f) vs saliency-ratio (L_q/L_d) design space with crash-safe JSONL result storage
• GPU-accelerated harmonic decomposition — CuPy FFT on CUDA (optional, falls back to NumPy)

Drive comparison

The demo runs three PMSM designs under the same current-vector controller (540 V DC bus, 20 A limit, 50 Hz electrical speed reference) with a 3 Nm load step at t = 1.2 s:

• A: 4-pole SPMSM — non-salient baseline (L_d = L_q = 4 mH)
• B: 4-pole IPMSM — saliency adds reluctance torque (L_q/L_d = 1.5), producing higher peak torque but drawing more current during acceleration
• C: 8-pole SPMSM — lower flux linkage (psi_f = 0.08 Wb) but higher torque-per-amp from doubled pole count, so less current under load at the cost of slower transient recovery

Verification and validation

Solver verification (Zhu & Howe)

The NGSolve FEM solver is verified against the Zhu & Howe (1993) closed-form air-gap field solution. The fundamental agrees within 0.2–0.3% across all configurations, confirming that the code solves the magnetostatic equations correctly.

Model validation (CREATOR benchmark)

Cross-checked against the CREATOR open-benchmark PMSM (70 W, 4-pole, 6-slot, sintered ferrite — Dhakal et al., COMPEL 2025):

Test	Computed	Published	Error
Back-EMF (E_0)	47.92 V	47.37 V	1.2%
Rated torque	0.102 Nm	0.100 Nm	1.9%
Max torque (at MTPA)	0.159 Nm	0.150 Nm	5.9%
FEM fundamental (B_1)	0.0200 T	0.0200 T	0.2%

The model captures the fundamental accurately. Higher harmonics (5th: 1.7% computed vs 13.2% measured) are underpredicted due to the sinusoidal magnetization assumption — real ferrite magnets have near-rectangular magnetization profiles that produce trapezoidal back-EMF. This is a quantified model limitation, not a solver error.

Slot count sensitivity

A sweep over Q = {6, 9, 12, 18, 24, 36} for a 4-pole SPMSM shows the trade-off between winding complexity and field quality:

Q=6:  THD ~ 12%  (wide slots, high distortion)
Q=12: THD ~  4%  (baseline)
Q=36: THD ~  1%  (narrow slots, smoother field)

Quick start

git clone https://github.com/jonelay/phase-sweep.git
cd phase-sweep
python -m venv .venv
.venv/bin/pip install -e .

# Optional: GPU support (requires CUDA 12.x)
.venv/bin/pip install -e ".[gpu]"

# Run the demo (generates 8 plots)
.venv/bin/python demo.py

# Run tests
.venv/bin/python -m pytest -v tests/

Or with uv:

uv venv .venv --python 3.12
uv pip install -e .
.venv/bin/python demo.py

Project structure

phasesweep/
├── configs.py        # Motor configurations (MotorConfig TypedDict) and constants
├── sim.py            # Drive simulation, FEM orchestration, sweeps
├── plots.py          # All plotting functions
├── fem_field.py      # NGSolve 2D FEM solver + CuPy harmonic decomposition
├── sweep_types.py    # MotorSweepConfig (frozen, MD5 config_id), SweepResult
├── result_store.py   # JSONL append-only store with JSON index, crash-safe resume
├── sim_runner.py     # Subprocess runner for motulator (60s timeout)
└── fem_runner.py     # Subprocess runner for NGSolve (300s timeout)
demo.py               # Entry point — runs comparisons, sweeps, and FEM analysis
tests/                # pytest integration tests (98 tests)
data/                 # CREATOR reference data (CC BY-NC-ND 4.0)
motors/               # Motor parameter files (TOML)

Dependencies

Package	Purpose	License
motulator	Synchronous machine drive simulation	MIT
NGSolve	2D finite element field solver	LGPL-2.1
NumPy	Numerical computation	BSD-3-Clause
Matplotlib	Plotting	PSF (BSD-compatible)
CuPy (optional)	GPU-accelerated harmonic decomposition	MIT

Reference data

The data/creator_case_pmsm/ directory contains reference data from the CREATOR open-benchmark PMSM (Dhakal et al., 2025), licensed under CC BY-NC-ND 4.0. See data/creator_case_pmsm/README.md for full attribution and citation info.

License

Code is licensed under the MIT License.

Reference data in data/ is licensed under CC BY-NC-ND 4.0 — see the data directory README for details.