MeshGuard Aerospace

MeshGuard Aerospace is my local-first tool for reviewing and documenting finite-element mesh-refinement studies. It keeps the study data, criteria, units and calculation trace together so the numerical evidence can be checked without relying on a hidden score.

MeshGuard assesses mesh-refinement evidence and numerical behaviour. It does not solve or physically validate an FEA model, verify loads or boundary conditions, establish design safety or provide certification evidence.

What it does

• accepts manual study data, structured JSON and quoted CSV imports;
• evaluates adjacent-level convergence for selected scalar quantities;
• checks force and moment reaction equilibrium separately;
• reports met, not_met and not_assessed with reasons and formula substitutions;
• identifies a trend consistent with a possible local stress singularity as a review warning, not proof;
• applies the documented equal-ratio three-grid Richardson and GCI method only when its gates pass;
• recommends one of the supplied mesh levels when the configured evidence supports it;
• converts SI and US customary project data with matching dimensional tolerances; and
• exports printable HTML, normalised CSV and canonical JSON evidence.

Run on Windows

1. Install Node.js 20 or later.
2. Double-click START_MESHGUARD.cmd.
3. Open http://127.0.0.1:4173 if the browser does not open automatically.
4. Keep the command window open while using the application.

The release includes its browser and build runtime files, so a separate npm install is not required.

Run from a terminal

npm run start:source

On Linux or macOS:

./start.sh

Verification

Run the core compiler, lint, build, smoke, unit, integration and independent numerical checks with:

npm run verify

Browser workflow and accessibility checks run separately:

npm run test:e2e
npm run test:a11y

These browser checks require Python 3 and Playwright. See VERIFICATION_REPORT.md for the recorded results and evidence boundary.

Engineering rules

The full method is documented in docs/contracts/CALCULATION_REFERENCE.md. Key rules include:

• relative change is |q_f-q_c|/|q_f| and is not assessed when q_f = 0;
• equality with a configured criterion passes;
• a recommendation needs at least three levels and at least one required quantity;
• every required transition from a candidate through the finest level must pass;
• force and moment equilibrium remain separate dimensional checks;
• GCI is not assessed when its documented applicability gates fail; and
• displayed rounding never changes the underlying decision.

Project structure

src/domain/        engineering calculations, validation, CSV and units
src/data/          synthetic demonstration studies
src/app.tsx        application and report workflow
tests/             calculation, workflow and browser checks
samples/           CSV, JSON and expected-result fixtures
docs/contracts/    engineering requirements and methods
scripts/           build, serve, screenshot and verification tools
screenshots/       production application views

Privacy

The application has no backend, login, analytics or telemetry. Projects are saved in browser storage. Clearing site data removes them, so JSON export should be used for backup. Exported hashes detect input changes but are not approval signatures.

Limitations

• MeshGuard accepts scalar study data and does not parse native solver databases.
• The possible-singularity rule is diagnostic only.
• GCI is limited to the documented monotonic equal-ratio three-grid method.
• Solver-specific quality thresholds must be supplied by the analyst.
• A recommended mesh is the coarsest supplied level that meets the configured numerical evidence from that point through the finest level. It is not proof of physical correctness.

See docs/KNOWN_LIMITATIONS.md for the complete list.

Licence and notices

Project source is supplied under LICENSE. Third-party components retain their own licences; see THIRD_PARTY_NOTICES.md.