Hugging Face's logo

Datasets:
bsebench-org
/
severson-2019-raw

Tasks:
Time Series Forecasting
Tabular Regression
Languages:
English
Size:
1B<n<10B
Tags:
battery
state-of-charge
state-of-health
lithium-ion
LFP
lithium-iron-phosphate
License:
Dataset card Files
xet
 Community
Dataset Viewer

The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

severson-2019-raw

Tier 1 raw mirror of the Severson et al. 2019 commercial LFP fastcharge cycling dataset, hosted under the BSEBench organization on the HuggingFace Hub. The files in this repository are preserved bit-exact as published on the original Toyota Research Institute data portal at data.matr.io. No values are modified; no columns are renamed; no rows are dropped. Every file's SHA-256 digest is recorded in the BSEBench manifest YAML and matches the original distribution.

This repository exists for provenance verification and audits only. For the BSEBench-canonical Parquet harmonization that consumers actually use for filter benchmarking, see the Tier 2 sibling repository bsebench-org/severson-2019.

Status

This is a placeholder card. The raw .mat files are not yet uploaded to the HuggingFace Hub. The planned upload pipeline is :

  1. Manual download of the three Severson batches from https://data.matr.io/1/projects/5c48dd2bc625d700019f3204 (registration may be required by the TRI portal).
  2. Local SHA-256 computation via scripts/upload_tier1_to_hf.py --src ./local --repo-id bsebench-org/severson-2019-raw --private --dry-run.
  3. Inventory cross-check against bsebench-datasets/manifests/severson_2019_lfp.yaml (committed only after real digests are populated — no fake checksums in this repository).
  4. Public upload (--dry-run removed) once the manifest validates.
  5. Update of this card with the populated ## File inventory section, the manifest commit SHA, and a verified_at timestamp.

Until step 5 is reached, treat the file inventory below as a best-effort estimate based on community references (BatteryML, BEEP, MIT Braatz Group GitHub repository), not as a directly verified manifest of HuggingFace content.

What this is

A bit-exact mirror of the dataset published with :

Severson, K. A., Attia, P. M., Jin, N., Perkins, N., Jiang, B., Yang, Z., Chen, M. H., Aykol, M., Herring, P. K., Fraggedakis, D., Bazant, M. Z., Harris, S. J., Chueh, W. C., Braatz, R. D. (2019). "Data-driven prediction of battery cycle life before capacity degradation." Nature Energy, 4(5), 383–391. doi:10.1038/s41560-019-0356-8

@article{severson2019datadriven,
  author  = {Severson, Kristen A. and Attia, Peter M. and Jin, Norman
             and Perkins, Nicholas and Jiang, Benben and Yang, Zi
             and Chen, Michael H. and Aykol, Muratahan
             and Herring, Patrick K. and Fraggedakis, Dimitrios
             and Bazant, Martin Z. and Harris, Stephen J.
             and Chueh, William C. and Braatz, Richard D.},
  title   = {Data-driven prediction of battery cycle life before
             capacity degradation},
  journal = {Nature Energy},
  volume  = {4},
  number  = {5},
  pages   = {383--391},
  year    = {2019},
  doi     = {10.1038/s41560-019-0356-8},
  url     = {https://www.nature.com/articles/s41560-019-0356-8},
}

Cell specifications

Property Value
Manufacturer A123 Systems
Model APR18650M1A
Form factor 18650 cylindrical
Cathode chemistry LFP (lithium iron phosphate, LiFePO4)
Anode chemistry Graphite
Nominal capacity 1.1 Ah
Nominal voltage 3.3 V
Charge cutoff (used) 3.6 V
Discharge cutoff (used) 2.0 V
Number of cells (this dataset) 124
End-of-life threshold 80 % capacity retention

Cycling protocol

All cells were cycled inside a 30 °C controlled environmental chamber. Charging was performed under one-step or two-step fast-charging policies spanning charge rates from 1C to 6C (corresponding to 8 to 13.3 minutes to reach 80 % SOC), giving a total of 72 distinct fast-charging strategies across the cohort. Discharging was uniform : 4C constant-current to the discharge cutoff. A 1-minute rest was enforced after reaching 80 % SOC during charging, and a 1-second rest after each discharge. Internal resistance was probed once per cycle by 10 pulses of ±3.6C with a pulse width of 30 or 33 ms.

This protocol is what makes Severson 2019 a strong stress-test for filter benchmarks : the cell-to-cell variation is dominated by charging policy rather than ambient conditions, isolating the protocol-driven aging mechanisms that filters are typically asked to compensate for.

File inventory (best-effort)

Severson 2019 is distributed as three .mat files (HDF5 v7.3 format) on the TRI data portal :

File Date Cells Size (approx.)
2017-05-12_batchdata_updated_struct_errorcorrect.mat 2017-05-12 46 2.82 GB
2017-06-30_batchdata_updated_struct_errorcorrect.mat 2017-06-30 48 1.80 GB
2018-04-12_batchdata_updated_struct_errorcorrect.mat 2018-04-12 46 3.01 GB
Total 140 channels → 124 cells after exclusions ~7.6 GB

A fourth file dated 2019-01-24 is sometimes seen in the same data.matr.io project ; that file belongs to Attia et al. 2020 ("Closed-loop optimization of fast-charging protocols for batteries with machine learning") and is not part of Severson 2019. This Tier 1 mirror covers only the three Severson 2019 batches.

The 16 channels that account for the gap between the 140 raw channels and the published cohort of 124 cells are documented in the upstream Braatz Group Load Data.ipynb notebook : five cells in batch 1 did not reach the 80 % capacity threshold (b1c8, b1c10, b1c12, b1c13, b1c22), five cells in batch 2 were re-assigned to batch 1 because they were continued from the first experimental run (b2c7, b2c8, b2c9, b2c15, b2c16), and six cells in batch 3 were excluded as noisy channels (b3c37, b3c2, b3c23, b3c32, b3c42, b3c43). The Tier 1 mirror still preserves these channels in the raw .mat files ; the Tier 2 canonical Parquet repository will apply the published exclusion mask.

Sizes are rounded community estimates (see BatteryML and the BatteryBits "Comparison of Open Datasets for Lithium-ion Battery Testing" article). Exact bytes will be locked once the actual upload to HuggingFace completes and manifests/severson_2019_lfp.yaml is populated with SHA-256 digests.

Why "raw mirror" tier

BSEBench follows a dual-tier dataset strategy :

  • Tier 1 (this repository) — the original .mat files, preserved byte-for-byte, with SHA-256 digests recorded in our manifest and cross-checked against the original publication's distribution. Use this tier if you need to verify provenance, run independent harmonizations, or audit our adapter's correctness.
  • Tier 2 — the BSEBench-canonical Parquet harmonization at bsebench-org/severson-2019. Consistent column names, BPX-1.1 sign convention, unified schema across all benchmark datasets. Use this tier for filter benchmarking and most downstream work.

Original source

The original Severson 2019 dataset was distributed via data.matr.io (the Toyota Research Institute Experimental Data Platform), specifically project 5c48dd2bc625d700019f3204. This URL is recorded as citation and provenance metadata only.

The HuggingFace Hub mirror at this repository is the BSEBench single source of truth for fetching. Adapters in bsebench-datasets never hit data.matr.io at runtime. This insulates the benchmark from upstream availability changes (URL shifts, registration requirements, bandwidth limits, eventual portal retirement) while preserving the citation chain back to the original publishers.

License

The Severson 2019 dataset is distributed under the Creative Commons Attribution 4.0 International (CC-BY-4.0) license, consistent with the licensing policy of the data.matr.io platform's earlier (pre-2025) datasets per the TRI Energy & Materials Datasets catalog.

Verbatim core grant from the CC-BY-4.0 legal code :

"Subject to the terms and conditions of this Public License, the Licensor hereby grants You a worldwide, royalty-free, non-sublicensable, non-exclusive, irrevocable license to exercise the Licensed Rights in the Licensed Material to: (1) reproduce and Share the Licensed Material, in whole or in part; and (2) produce, reproduce, and Share Adapted Material."

The redistribution rights granted by this license are the legal basis on which BSEBench mirrors the dataset on the HuggingFace Hub. Attribution is given to the original authors via the BibTeX block above and via the manifest's citation_bibtex field. Derivative material (the Tier 2 Parquet harmonization at bsebench-org/severson-2019) is offered under the same CC-BY-4.0 license, with BSEBench attribution added on top of the original Severson 2019 attribution chain.

Note : the publication text of the Nature Energy paper is governed by Springer-Nature's text-and-data-mining terms (CrossRef license type tdm, effective 2019-03-25), which is a separate licensing regime from the dataset hosted on data.matr.io. CC-BY-4.0 covers the experimental data only ; do not assume it covers the paper PDF.

How to use 

from huggingface_hub import snapshot_download

local_dir = snapshot_download(
    "bsebench-org/severson-2019-raw",
    repo_type="dataset",
)
# local_dir contains the three .mat files, SHA-256 verified
# against bsebench-datasets/manifests/severson_2019_lfp.yaml

To then read a .mat file in Python (the files are HDF5 v7.3, not classic v5, so scipy.io.loadmat will not work — use h5py) :

import h5py
from pathlib import Path

p = Path(local_dir) / "2017-05-12_batchdata_updated_struct_errorcorrect.mat"
with h5py.File(p, "r") as f:
    print(list(f.keys()))                  # ['#refs#', '#subsystem#', 'batch', 'batch_date']
    batch = f["batch"]
    print(list(batch.keys()))               # ['Vdlin', 'barcode', 'channel_id',
                                            #  'cycle_life', 'cycles', 'policy',
                                            #  'policy_readable', 'summary']

For the BSEBench-harmonized Parquet version that exposes a clean benchmark-ready API, prefer bsebench-org/severson-2019.

Citation

Cite the original Severson 2019 paper (BibTeX above). BSEBench's contribution is hosting and harmonization, not the data itself.

If you also use BSEBench tooling for filter benchmarking, additionally cite :

@misc{bsebench2026,
  author = {Akir, Oussama and {BSEBench Contributors}},
  title  = {{BSEBench}: an open-source benchmark for battery
            state-estimation filters},
  year   = {2026},
  url    = {https://bsebench.org},
}

Provenance manifest

A machine-readable manifest validating this dataset's metadata against the bsebench-dataset-manifest/v1 Pydantic v2 schema lives at bsebench-org/bsebench-datasets/manifests/severson_2019_lfp.yaml.

The manifest records, for every .mat file in this repository :

  • source.canonical_url — the data.matr.io project URL
  • source.canonical_doi — the Nature Energy DOI for citation
  • source.publication_authors and source.publication_year
  • per-file path, sha256, and size_bytes
  • the dataset-wide license (SPDX CC-BY-4.0) and redistribution_allowed flag (true)
  • the citation_bibtex block (verbatim copy of the BibTeX above)
  • huggingface_tier1_repo (= bsebench-org/severson-2019-raw) and huggingface_tier2_repo (= bsebench-org/severson-2019)

The manifest is committed only after the SHA-256 digests are populated from the actual HuggingFace mirror — never with placeholder values.

See also

Downloads last month
5
Total file size:
8.27 GB