---
title: FELA-PdM playground
colorFrom: yellow
colorTo: red
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: other
---

# FELA-PdM playground

Feed a sensor window and see the model's call. Two tabs: remaining useful life (C-MAPSS, 30
cycles of 14 sensors) and bearing fault (CWRU, 2048 vibration samples). Runs on CPU.

## The demo data (and why it shows generalization)

The remaining useful life head was trained on C-MAPSS subset FD001 (a single operating
condition). The other subsets, FD002, FD003, and FD004, use different operating conditions and
fault modes and were not in this head's training set, so a window from those subsets is a real
out of distribution test. They are public NASA data from the Prognostics Data Repository
(Saxena et al. 2008); download a window and paste it into the RUL tab.

The bundled examples (the synthetic RUL window and the synthetic vibration window) are clearly
labeled as synthetic and illustrative. They let the Space load fast without a large download.
They are not real measurements and should not be read as benchmark results.

## Weights

Set `FELA_PDM_WEIGHTS` (or a Space secret) to a directory holding the per head safetensors
(`cmapss_FD001.safetensors`, `cwru.safetensors`) and `config.json`. The weights are in the
Hugging Face repo `lowdown-labs/FELA-pdm`.

## Note

For research and illustration only. This is not a safety critical controller. Do not act on the
remaining useful life number without independent validation against your own field data.