The Prognostics, Health Management (PHM) and Forecasting Benchmark Arena

Getting started

Full documentation is available in docs/.

Quickstart

Prerequisites

• Python 3.12+
• uv
• Git with submodule support

Install

Standard (with foundation models, from git)

Installs TabPFN, TabDPT, carte-ai and phmd directly from their git repositories.

git clone <repo-url>
cd picid
uv sync

Without foundation models

Lighter install — skips the TabFM and phmd groups entirely.

git clone <repo-url>
cd picid
uv sync --no-group tabfm --no-group phmd

phmd only (no foundation models)

Installs phmd from git but skips TabPFN / TabDPT / carte-ai.

git clone <repo-url>
cd picid
uv sync --no-group tabfm

Local editable install (for developing foundation models)

Uses editable installs from local checkouts after the standard git-backed sync.

git clone <repo-url>
cd picid
bash scripts/sync_local_packages.sh

Configure paths

Copy configs/paths/default.yaml to a new file (e.g. configs/paths/local.yaml) and set:

root_dir: /path/to/project
data_dir: /path/to/data
output_dir: /path/to/outputs
cache_path: /path/to/cache

Datasets

Most datasets are downloaded automatically the first time the datasource is executed, if the data is not already present locally. This covers: UNIBO21, NB14, PHME20, PRONOSTIA, XJTU-SY, HSF-15, N-CMAPSS (phmd variant), and ThreeW.

Note: The N-CMAPSS (concepts variant) datasource does not yet have automatic download support and requires manual placement of the raw NASA HDF5 files. This will be added in a future update.

Run your first experiment

uv run picid-run paths=local debug=default experiment=unibo/prognostics/raw/cnn_1d

Or by activating the venv first:

source .venv/bin/activate
python picid/run.py paths=local debug=default experiment=unibo/prognostics/raw/cnn_1d

For full configuration options see the documentation.

Documentation

To browse the full docs locally:

uv run mkdocs serve

Then open http://127.0.0.1:8000.

Citation

Several works have been implemented in Picid. If you wish to cite us please use:

• Picid: A Modular Evaluation Infrastructure for Reproducible PHM Across Tasks and Domains (05/2026)
• From paper to benchmark: agentic, framework-based reproduction of under-specified methods in machine health intelligence (05/2026)
• Towards Unified and Data-Efficient Prognostics and Health Management with Tabular Foundation Models (upcoming!)

@article{telyatnikov2026picid,
  title={Picid: A Modular Evaluation Infrastructure for Reproducible PHM Across Tasks and Domains},
  author={Telyatnikov, Lev and Theiler, Raffael and Von Krannichfeldt, Leandro and Fink, Olga},
  journal={arXiv preprint arXiv:2605.28345},
  year={2026}
}

@article{theiler2026paper,
  title={From paper to benchmark: agentic, framework-based reproduction of under-specified methods in machine health intelligence},
  author={Theiler, Raffael and Comito, Ludovico and Leko, David and Von Krannichfeldt, Leandro and Telyatnikov, Lev and Fink, Olga},
  journal={arXiv preprint arXiv:2605.28371},
  year={2026}
}

License

This project is licensed under a non-commercial license. Use is permitted for academic and government-funded research only. Commercial use is prohibited. See LICENSE.txt for full terms.