Stope Shape Optimizer

Stope Shape Optimizer scans mining block models for economic underground stope shapes, visualizes them in 3D, and exports to AutoCAD DXF, CSV, PDF & Excel — built with FastAPI & React.

Comment

Inspiration

Mining planners still spend hours working through block model CSVs in spreadsheets, with little real-time 3D feedback and slow handoffs to AutoCAD. I wanted to build something real — not another generic hackathon app — that modernizes a genuine industry workflow.

Underground stope design sits at the intersection of 3D spatial data, domain-specific math, and practical engineering outputs. That made it a perfect fit for a full-stack web app: fast APIs, in-browser 3D, and instant exports engineers actually use.

What it does

Stope Shape Optimizer turns a geological block model CSV into economic stope shapes you can explore and export.

  1. Upload a block model (coordinates, grade, density)
  2. Configure stope dimensions (30m × 20m × 5m) and a grade cutoff
  3. Scan the full 3D model for every economic stope position
  4. Visualize ore blocks in an interactive Three.js viewer, color-coded by grade
  5. Export to AutoCAD DXF, CSV, PDF, and Excel

On a ~97,000 block dataset, the scan completes in under 300 milliseconds and finds ~36,000 economic stopes at a 10 g/t cutoff.

For each stope window, the engine computes mass-weighted average grade:

[ \bar{G} = \frac{\sum_{i=1}^{n} T_i \cdot G_i}{\sum_{i=1}^{n} T_i} ]

where (T_i = V_i \times \rho_i) and a stope is economic when (\bar{G} \geq G_{cutoff}).

Missing blocks inside a stope are treated as waste rock ((G = 0) g/t, (\rho = 2.8) t/m³) so dilution is modelled correctly.

How we built it

Backend — Python / FastAPI

  • parser_csv.py — Parses CSV block models, skips metadata rows, auto-detects grade/density columns
  • optimizer.py — Builds a 3D NumPy grid, applies gap fill, scans all stope positions via sliding_window_view
  • dxf_exporter.py — Exports stope prisms as 3D faces for AutoCAD (ezdxf)
  • reporter.py — Generates CSV, PDF, and Excel reports

Frontend — React / Vite / Three.js

  • 4-step tab UI: Upload → Scenario → 3D View → Reports
  • Instanced Three.js meshes for rendering 47k+ grade-colored blocks
  • Tailwind CSS dark theme

Architecture

CSV Upload → FastAPI → NumPy 3D Grid → Stope Scanner → JSON Results
                                              ↓
                         React UI ← 3D Viewer + DXF / CSV / PDF / Excel

Challenges we ran into

Sparse block models — Empty grid cells were initially ignored, inflating stope grades. Gap fill (0 g/t @ 2.8 t/m³) fixed dilution modelling.

Performance — Scanning every stope position on ~97k blocks required vectorized NumPy sliding windows instead of Python loops — runtime dropped from seconds to ~300 ms.

3D at scale — Rendering tens of thousands of blocks required instanced meshes grouped by grade tier, not individual components per block.

DXF file size — Exporting 36k+ stope prisms with labels creates huge files. We export all geometry but skip text labels on large exports to keep AutoCAD usable.

Cross-platform setup — Windows and Mac/Linux setup scripts plus a clean Python venv workflow so anyone can run the demo locally.

Accomplishments that we're proud of

  • Built a domain-realistic mining tool, not a generic CRUD app
  • Sub-300ms stope scan on a real ~97k block dataset
  • End-to-end pipeline: CSV in → 3D visualization → AutoCAD DXF out
  • Correct mining formulas: volume, tonnes, weighted grade, gold content, ore vs waste
  • Polished UI with live stats, grade legends, opacity controls, and multi-format downloads
  • Gap-aware sliding-window algorithm that matches real mining engineering logic

What we learned

  • Domain knowledge matters — Gap fill, dilution, and weighted cutoffs are as important as the code
  • NumPy sliding_window_view is powerful for 3D spatial scanning problems
  • Three.js instancing is essential for large-scale in-browser 3D
  • Export formats are first-class features — engineers live in AutoCAD and Excel, not JSON
  • A simple 4-step UX makes complex engineering tools approachable for demos and judges

What's next for Stope Shape Optimizer

  • Adjustable stope dimensions in the UI (not just cutoff grade)
  • Economic NPV optimization — metal price, mining cost, processing cost, recovery
  • Best-set stope selection — deduplicate overlapping stopes and maximize total value
  • Deployed live demo — hosted version judges can try without local setup
  • Support for multiple metals — copper, zinc, multi-element block models
  • Improved 3D viewer — render stope prisms (not just ore blocks) with accurate block sizes

Built With

Share this project:

Updates

Submission history