Docs / Overview

cufemlab documentation

cufemlab™ is a proprietary GPU finite-element service from Secrotec B.V. for 2-D permanent-magnet-machine analysis. It is self-serve: sign in at /app, create a project, pick a workflow and run it in the browser. These docs cover the app’s workflow catalog, the result.v1 envelope, the cufem engine API, the Python cufemlab_client SDK, and the authoritative scope statement. Prefer we run a scoped case for you? An assisted pilot is also available.

Quickstart. Open the app → create an account (free-trial credits) → New analysis → pick a workflow → Submit → watch it run → download the signed report and artifacts. Or drive the same live API with the Python SDK (Client.from_env()) — see Getting started and Examples.

Current SaaS capability matrix

What you can run in the live app today. The cross-checked validated reference is moving-band cogging; operating-torque and iron-loss are honest engineering estimates, not the cross-checked reference and not third-party-certified.

Workflow	Status	Notes
cogging_torque_movingband	flagship · validated	Convergence-grade moving-band (Arkkio) cogging; cross-checked vs FEMM/GetDP with a per-run validation card.
pmsm_operating_torque	available · estimate	Steady-state torque at an operating point (GPU). Engineering estimate with a mesh-convergence check; not the cross-checked reference.
iron_loss_estimate	available · estimate	Approximate iron-loss from a reference field, split by region. An estimate for a given steel grade; not a certified core-loss figure, not a hysteresis/eddy mechanism split.
signed_report_generation	available	Deterministic signed report for a completed job; SHA-256 tamper-evidence (integrity hash, not a PKI signature).
material_comparison	available · free-eligible	Side-by-side comparison of cited material profiles.
demo_motor_quick_check	available · free-eligible	Fast smoke check for geometry and pipeline.
cogging_sweep_2d	legacy · superseded	Earlier 2-D angle-sweep cogging path, kept for continuity. Use moving-band cogging for the convergence-grade, cross-checked result.
thermal_conduction_steady	new · internally validated	GPU steady heat conduction (P1 tets); MMS + energy-balance validated. `validation_card=not_validated` (no external passport yet).
thermal_conduction_transient	new · internally validated	GPU transient heat conduction (θ-method, Crank-Nicolson order 2). Internally validated.
eddy_thermal_coupled	new · internally validated	Eddy skin-effect Joule loss driving a thermal response (energy chain).
core_loss_estimate	new · internally validated	Bertotti 3-term core-loss separation with cited coefficients (CPU).
coreloss_thermal_coupled	new · internally validated	Core-loss density driving a cooled-block thermal response.
field_to_field_neural_operator	new · experimental	In-distribution neural surrogate / self-validating demo (held-out 0.68% on the training distribution; OOD degrades). Not general-purpose; custom-field input is a follow-on.

API / SDK / product surface

Every workflow above is reachable through the browser app and the same live REST API. What the product surface exposes:

Auth & API keys — register / login, and an API-key lifecycle (create, list, revoke) for programmatic access.
Projects, files, jobs — create projects, upload geometry/parameter files, submit jobs, poll status, and read results.
result.v1 envelope — a canonical result payload: summary, metrics, artifacts, provenance and a validation card.
Artifacts & signed reports — authenticated downloads of result artifacts and the signed PDF report (bearer-token, not public links).
Credits & quota — free-trial credits for free-eligible workflows; paid credits are in beta (not yet production billing).
Python SDK — cufemlab_client with Client.from_env(), analyze(), job.wait(), job.result(), report_pdf(), download_artifact().
GPU Jupyter — per-session notebook containers with the cufem engine, for lower-level scripting.

Two lanes

Lane A · 2-D · validated

2-D PMSM moving-band cogging

The cross-checked validated reference — airgap flux plus a convergence-grade moving-band cogging torque, checked against the open-source FEMM and GetDP solvers with a per-run validation card. This is the production path in the app.

~0.048 Nm · cross-checked vs FEMM/GetDP · GPU

Lane B · cufem3d · 3-D field validation

3-D field validation & geometry pipeline

GPU-first 3-D PMSM geometry, meshing, rotor-sweep and magnetic field/flux evaluation with VTU field output — the 3-D validation lane. Production-grade 3-D torque, end-effects and skew are roadmap extensions on the same pipeline.

geometry/mesh ready · field/flux evidence · 3D torque roadmap

Live / Available with limits / Roadmap

Live in the app (self-serve)

Moving-band cogging torque — cross-checked, validated reference
PMSM operating-point torque — estimate
Iron-loss estimate; material comparison; quick check
Signed reports, artifacts, result.v1, API-key + SDK, GPU Jupyter
Thermal conduction (steady/transient), eddy→thermal & core-loss→thermal coupling, core-loss — internally validated (external passport pending)
Field-to-field neural surrogate — experimental, in-distribution

Available with stated limits

Operating-torque / iron-loss are estimates, not the cross-checked reference
Legacy 2-D cogging sweep — superseded by moving-band
3-D field validation & geometry/mesh pipeline; full 3-D torque is roadmap
Paid credits are in beta — not production billing

Roadmap / not claimed

Full production 3-D torque, end-effects, skew — roadmap
Time-harmonic & transient eddy currents, conductive regions — roadmap
External validation passports for the new thermal / core-loss types; general-purpose neural operator with custom-field input — R&D
Not a drop-in replacement for ANSYS Maxwell / COMSOL / JMAG

What we do not claim. The new thermal, core-loss and coupling types are validated internally (MMS / energy-balance / cited coefficients) but carry no external validation passport yet (validation_card=not_validated); the neural surrogate is an in-distribution self-validating demo, not a general-purpose operator, and does not accept arbitrary custom fields. We do not claim validated production 3-D torque, externally-certified iron / core-loss, standalone eddy-current as a customer API type, or any measured speed / accuracy advantage over a commercial tool (ANSYS Maxwell / JMAG / COMSOL have not been benchmarked). Two things are cross-checked against an external reference: the 2-D moving-band cogging workflow (FEMM / GetDP) and CPU 3-D field validation (an analytic sphere); operating-torque and iron-loss are engineering estimates.

We publish our negatives. The same pre-declared ±20% bar that validates the moving-band cogging path was applied to a first 3-D (2.5-D) torque attempt — and it did not pass; that result is recorded openly as a FAIL, not hidden. That is why cufem3d torque is not claimed.

Roadmap

These are roadmap / R&D, not current capabilities, and are not claimed as working today (the thermal, core-loss, coupling and neural-surrogate types that shipped 2026-07-11 are covered above as live, with their honest validation tiers):

Full production 3-D motor torque (end-effects, skew).
Standalone time-harmonic / transient eddy currents and conductive regions as customer API types (engine-internal today).
External validation passports for the live thermal / core-loss / coupling types; full 3-D thermal maps.
Full industrial core-loss package, full material database, operating-map validation.
General-purpose neural operator with arbitrary custom-field / CAD input (the live surrogate is in-distribution only).
External commercial benchmarking (ANSYS / JMAG / COMSOL) — not yet measured.

See the full staged plan on the roadmap page.

Start here

Getting started

Open the app, create an account, run a workflow, and read the result — or drive the same live API with the Python SDK.

Open the guide →

Motor workflow

The 2-D PMSM moving-band cogging workflow — airgap flux, convergence ladder, and the FEMM/GetDP cross-check on every run.

See the workflow →

Examples & recipes

Ready-to-run SDK snippets for the app’s workflows — submit, poll, read result.v1, and download the report.

Browse recipes →

cufem engine API

The real cufem GPU solver classes: the mesh, the magnetostatic problem, the result envelope, and the validation-card schema.

Open reference →

What cufemlab is

cufemlab is a self-serve GPU finite-element service for 2-D PMSM analysis — 2-D linear-µ magnetostatics, airgap flux and moving-band cogging on an RTX-class GPU, plus operating-torque and iron-loss estimates. Every cogging result is cross-checked against the open-source FEMM and GetDP solvers, cites a tolerance, reports its measured error, and carries a signed, deterministic evidence pack.

It is designed for:

Electric-motor R&D teams who need audit-grade 2-D PMSM cogging with an independent cross-check.
Teams that need a signed, reproducible evidence pack for a cogging result rather than “it ran.”
Research labs that compare numerical methods and need verdicts they can cite.
Coursework and method comparison where reproducibility matters.

Honest scope

The shipped product is a focused 2-D PMSM tool, not a general multiphysics suite; the cross-checked validated claim is the 2-D moving-band cogging workflow, and operating-torque and iron-loss are estimates. Separately, an internal 24-phase research V&V program backs the underlying numerical methods — it is method validation, not a list of product features. In that program all 24 roadmap phases are at least MATURING today, and 16 are fully DONE (A, B, C, D, F, G, H, K, L, M, Q, R, S, T, U, V — verdict PASS, zero NOT_IMPLEMENTED). The other 8 phases (E, I, J, N, O, P, W, X) are MATURING: passing benchmarks plus honestly-deferred items (e.g. E Re>0 CFD, N full battery cell, W multi-GPU/HPC, plus parts of X certification) on the public 24-phase roadmap with dated estimates. Read the hard limitations page before relying on cufemlab in production work — every line is authoritative.

Discipline

Every result in cufemlab follows ten platform rules — most importantly: no PASS without a cited benchmark, a declared tolerance, and a measured error. The integration runner exits non-zero on any FAIL, and every report carries a SHA-256 provenance hash. See the validation section on the homepage for the full suite list.

How you access cufemlab

Self-serve at /app. Create an account, get free-trial credits, and run the free-eligible workflows immediately in the browser; results come with a validation card, a signed report and downloadable artifacts. Paid credits are in beta (not yet production billing). Prefer we run a scoped case for you, or need on-prem? Request an assisted pilot or email info@secrotec.nl.

Engagement modes:

Self-serve app — sign in at /app, run workflows yourself, download signed reports and artifacts, use the API/SDK.
Assisted pilot — a scoped 2-D pilot on your geometry with pre-declared success criteria and an independent FEMM/GetDP cross-check.
On-prem / design-partner — for regulated industries or sensitive geometry; we deploy on a GPU node we provision (cloud) or your workstation (on-prem). See pilot tiers.

License & ownership

To request a license, evaluation, or pilot, write to info@secrotec.nl.