---
title: CMAPSS Reliability Dashboard
emoji: 🛩️
colorFrom: blue
colorTo: green
sdk: docker
app_port: 7860
pinned: false
license: mit
---

# CMAPSS Reliability & RUL Dashboard

Interactive dashboard for the NASA CMAPSS turbofan-degradation dataset
(subsets FD001, FD002, FD004). For each subset:

- Operational-regime clustering (KMeans, k=1 for FD001, k=6 for FD002/FD004)
- Per-regime sensor normalization
- Weibull fleet model (β, η, MTTF), with 2-component mixture overlay on FD004
- CatBoost RUL prediction (piecewise target capped at 150 cycles)

## Local run with uv

```bash
uv sync
uv run python app.py
```

## Files

- `app.py` — Gradio dashboard
- `rul_multi.py` — feature engineering, regime clustering, normalization
- `weibull_multi.py` — 2-component Weibull EM mixture fit
- `data/` — 9 CMAPSS files (train/test/RUL × 3 subsets)
- `cache/` — pre-trained CatBoost models, one per subset
- `Dockerfile` — uv-based image for HF Spaces

## Deployment to Hugging Face Spaces

1. Generate the lockfile locally (one-shot):

   ```bash
   uv lock
   ```

2. Create a new Space on huggingface.co (SDK = Docker), clone it, copy
   every file from this directory into the clone, then push:

   ```bash
   git lfs install
   git lfs track "data/*.txt" "cache/*.cbm"
   git add .gitattributes
   git add .
   git commit -m "Initial deploy"
   git push
   ```

   The `.txt` and `.cbm` files are tracked via Git LFS to keep the repo
   light. The Space will build the Docker image and start the app on
   port 7860.

## Model card

| Subset | Weibull β | Weibull η | Test RMSE | Test NASA |
|---|---|---|---|---|
| FD001 | 4.55 | 225 | ~16.7 | ~570 |
| FD002 | 4.48 | 229 | ~27.0 | ~8 600 |
| FD004 | 3.23 | 278 | ~27.6 | ~7 800 |