femrep — femis-governed FEM report generator

Turns Ansys / Nastran result files (.rst, .rth, .f06) into defensible, standardized engineering reports (PDF + DOCX). Reporter only — it ingests results your solve scripts already produce; it never drives the solver.

Governed by femis (engineering claim discipline, provenance, gates, no autonomous sign-off) and disciplined by ponytail (minimalism / YAGNI).

Why it exists

A "beautiful" CAE report that omits provenance, gate verdicts, and honest claim phrasing is worse than a plain one — it looks authoritative while being untrustworthy. femrep is the thin layer that makes a report defensible, not just pretty: every number is tied to its deck + result via a run manifest, phrased through public report language, and backed by a gates checklist that is computed — never invented to "pass."

Supported backends

Format	Backend	Notes
Ansys .rst	DPF	structural: von-Mises stress, displacement, contour
Ansys .rth	DPF	thermal: temperature; multi-set transient sequences
Nastran .f06	stdlib parser	SOL 153/159 thermal, transient + steady
Nastran .op2	pyNastran (not wired yet)	fails loudly with a .f06 companion hint

Quick start

# install (editable, from the repo)
pip install -e ".[gui]"

# CLI — any backend
femrep path/to/result.rst --log solve.mntr --out report.pdf --html --package
femrep path/to/result.f06 --out report.docx --template client
femrep doctor

# project/run library
femrep init DemoProject
femrep path/to/result.f06 --project femrep_projects/DemoProject --run-name run001 --html --package

# mesh convergence wizard: CSV with h,f or h,result columns
femrep gci grids.csv --out gci_runs.json --qoi peak_temperature_K

# batch mode: JSON with {"runs":[{"result":"...","out":"..."}]}
femrep batch runs.json

# bundled demo data
femrep examples/tiny_thermal.f06 --out demo_report.pdf --html --package
femrep gci examples/gci_points.csv --out demo_gci_runs.json --qoi demo_temperature
femrep batch examples/runs.json

# custom report template (branding + section layout)
femrep path/to/result.f06 --out report.pdf --template-file my_company.json

# desktop GUI (build + reuse templates visually)
femrep-gui

Report templates

A template is a reusable report definition — branding / title block (logo, company, customer, document number, colours, fonts) plus the section layout (which of the nine sections appear, in what order, with optional per-section intro text). Templates are per-project JSON files under <project>/templates/<name>.json.

Build and reuse them in the GUI: open a project, then Manage templates… to create one from scratch or seed it from a result file, tick/reorder sections, fill in your branding, and Save. Pick it from the Template dropdown before Generate report. From the CLI, pass --template-file <name>.json. With no template, the full nine-section report renders exactly as before.

Russian GOST report (ГОСТ 7.32-2017)

A template can set profile: "gost_ru" to emit a fully Russian отчёт о НИР as DOCX, structured and formatted per ГОСТ 7.32-2017 (титульный лист, реферат, содержание, введение, нумерованная основная часть, заключение; Times New Roman 14, полуторный интервал, поля 30/15/20/20 мм, сквозная нумерация). The реферат, введение, and заключение are auto-generated from the results to minimize manual editing. Select Шаблон → ГОСТ 7.32-2017 (DOCX, рус.) in the GUI (no project needed), or:

femrep result.f06 --out отчёт.docx --profile gost_ru

The profile uses extra title-page fields (организация, УДК, город, год, руководители) editable in the template. The English default profile is unaffected.

Both renderers (PDF via reportlab, DOCX via python-docx) emit the same content: cover, summary, readiness, model, meshing, composites/CFRP, solve + convergence, results (contour + time-history figures), mesh-independence (GCI), governance (gates + manifest). The optional HTML review includes the evidence dashboard, figure gallery, and a rotatable PyVista contour viewer when geometry is available.

Architecture

result file (.rst/.rth/.f06)
      │  backends/ (one adapter per format — same results.json schema)
      ▼
   extract.py        → results.json
      │
      ▼  govern.py (femis): manifest + claim phrasing + gates + GCI + readiness
   manifest.json + checks.json
      │
      ▼  figures.py: pyvista contour + matplotlib time-history
   *.png
      │
      ▼  report_pdf.py / report_docx.py
   report.pdf / report.docx

Each layer reads the previous layer's JSON and is runnable standalone — the repo's incremental stage() philosophy.

femis governance, enforced

• Issued-report language on the cover; internal execution modes stay out of customer-facing PDFs/DOCX files.
• Gates checklist — units, connectivity, equilibrium/heat balance, convergence, singularity, mesh-independence (GCI) — each computed, never invented.
• Evidence readiness — complete/missing/blocked evidence is summarized before export so limitations stay visible.
• Run manifest — solver + version, deck + result SHA-256, command line, units, mesh, material sources. NAFEMS R0033 traceability spine.
• No autonomous sign-off — femrep assembles evidence; a qualified engineer accepts it.

Composites / CFRP

No ACP/CFRP data ships with the project. The composite section renders a synthetic [0/90/0] T300/5208 CLT validation case (cases/clt_synthetic.py) — ABD matrix, B≈0 (symmetric), Tsai-Wu first-ply-failure — so the section's governance (failure philosophy, mesh-objectivity regularization, as-draped angles) is demonstrated. Real ACP/.rmed data is wired when supplied.

Honest limitations

• Reporter only — no solver orchestration. GCI ingests 3 pre-run results.
• .op2 is intentionally blocked until the binary adapter is wired. Use .f06; if a same-stem companion exists, femrep points to it instead of rendering an empty report.
• Convergence is read from Ansys .mntr (substep history) or .f06 output tables. The corpus is linear thermal — no Newton-Raphson residual history, and the claim phrasing says so honestly.
• Distributed .rst — point femrep at the consolidated file.rst (e.g. in a Workbench ..._files/dp0/.../MECH/ folder). The per-domain file0.rst..fileN.rst written by a multi-core solve are not time steps; femrep reads the consolidated file directly and never expands those domains. Numbered transient sequences are an Ansys thermal (.rth) concept only.
• DPF transport — femrep runs DPF locally against your own Ansys install and forces an in-process / unsecured-gRPC channel (DPF_GRPC_MODE=insecure), so a single-machine run never needs the mutual-TLS certificates that ansys-dpf-core ≥ 0.15 (Ansys 2026 R1) require by default. Override the env var if you connect to a secured remote DPF server.

License

Apache-2.0.