nasa-rw-2014-raw

Tier 1 raw mirror of the NASA Prognostics Center of Excellence (PCoE) Randomized Battery Usage Data Set (2014), hosted under the BSEBench organization for reproducibility purposes. Files are preserved exactly as published by NASA Ames.

Disambiguation — read this first

This repository mirrors the 2014 Randomized Battery Usage dataset, distributed by the NASA PCoE Data Repository as item #11 (Randomized Battery Usage Data Set). It is not the older 2007–2008 Battery Aging ARC FY08Q4 dataset (cells B0005, B0006, B0007, B0018, etc., authored by B. Saha and K. Goebel) which is item #5 in the same repository. The 2007 B-series uses constant-current discharge profiles sampled at a coarser rate and is excluded from BSEBench v1.0 because its sampling cadence and cycling protocol are not informative for the filter-divergence stress tests that motivate BSEBench's selection criteria.

If you are looking for B0005.mat etc., this is the wrong repository. The 2007 B-series is not currently mirrored under BSEBench.

Status

🚧 Upload pending — placeholder dataset card. The raw .mat files will be uploaded once verified for SHA-256 integrity against the original ZIP archive distributed by the NASA PCoE Data Repository (11.+Randomized+Battery+Usage+Data+Set.zip).

What this is

A bit-exact mirror of the dataset distributed under the entry "Randomized Battery Usage Data Set" on the NASA PCoE Data Repository, authored by B. Bole, C. Kulkarni, and M. Daigle at NASA Ames Research Center, recorded in 2014.

Suggested citation (NASA PCoE record) :

@misc{bole2014_nasa_rw,
  author    = {Bole, Brian and Kulkarni, Chetan S. and Daigle, Matthew},
  title     = {Randomized Battery Usage Data Set},
  year      = {2014},
  publisher = {NASA Ames Prognostics Center of Excellence (PCoE)},
  howpublished = {NASA PCoE Data Repository, Item~\#11},
  url       = {https://www.nasa.gov/intelligent-systems-division/discovery-and-systems-health/pcoe/pcoe-data-set-repository/}
}

The dataset comprises 28 commercial 18650 Li-ion cells with lithium cobalt oxide (LCO) cathodes and a nominal capacity of approximately 2.1 Ah, organized into seven groups of four cells each. Five of the seven groups were cycled at room temperature ; the remaining two groups were cycled at 40 °C. Each group is exposed to a distinct Random Walk (RW) current profile drawn from a different probability distribution (uniform, right-skewed, left-skewed) — yielding the seven sub-parts named on the NASA PCoE landing page :

1. Random Walk (RW1–RW4 — ±4.5 A continuous random walk)
2. Room-Temperature Random Walk
3. Room-Temperature Variable Random Walk
4. 40 °C Right-Skewed Random Walk
5. High-Temperature Right-Skewed Random Walk
6. 40 °C Left-Skewed Random Walk
7. Low-Temperature Left-Skewed Random Walk

(The exact part-to-cell-ID mapping is documented per-.mat-file inside each archive ; cell IDs follow the convention RW1 through RW28.)

The cycling protocol alternates two regimes : (a) Random Walk discharge/charge, in which the current setpoint is updated every 5 minutes to a value drawn at random from the group's discharge distribution table (e.g. {0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4} A for the uniform group) ; and (b) pulsed-load characterization cycles, run after every set of 50 random-walk cycles, providing reference capacity and EIS-impedance measurements that anchor health estimates.

Significance for BSEBench

The Random Walk profile is the single most informative stress test for state-estimation filter robustness in the entire NASA PCoE catalog. Three reasons :

1. Persistent excitation across the SOC band. Most public datasets exercise the cell with constant or piecewise-constant currents, which leaves the observability Gramian of an equivalent-circuit-model state estimator nearly degenerate — the filter's covariance collapses along the eigendirection aligned with the (slow) SOC mode, and on-line residuals stop being informative. The RW profile injects broadband current excitation by construction, exposing diagnostic pathologies (covariance collapse, hyper-confident divergence) that constant-current benchmarks hide.
2. Realistic automotive-like transients without the cost of HPPC. The 5-minute cadence and skewed amplitude distributions emulate the spectrum of currents seen in real automotive duty cycles without requiring the bench-engineering effort of replaying full FUDS / US06 profiles.
3. Cross-temperature span. Five room-temperature and two 40 °C groups (plus low-temperature variants) let users measure how filter covariance evolution depends on Arrhenius-driven internal-resistance shifts.

For these reasons, NASA RW 2014 is a mandatory benchmark for any Joint UKF, Joint EKF, particle filter, or moving-horizon estimator that claims robustness to dynamic load.

Cell specifications (verified)

Property	Value	Confidence
Cell count	28	HIGH
Form factor	18650 cylindrical	HIGH
Cathode chemistry	Lithium cobalt oxide (LCO)	HIGH
Nominal capacity	~2.1 Ah	HIGH (some sources state 2.2 Ah ; cell-specific value to be re-verified per cell from .mat headers)
Group count	7 (4 cells each)	HIGH
Temperature setpoints	5 groups @ room temperature, 2 groups @ 40 °C (plus low-temperature variant)	HIGH
Cycling protocol	Alternating Random Walk + pulsed-load reference cycles every 50 RW cycles	HIGH
Recording year	2014	HIGH

Cycling protocol

• Random Walk discharge : current setpoint updated every 5 min to a value randomly selected from the group's discharge distribution. For the uniform group, the alphabet is {0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0} A.
• Random Walk charge : groups RW9–RW12 are randomly charged AND discharged in a continuous random walk over [-4.5, +4.5] A. The remaining groups use a fixed CC-charge step to a target voltage followed by random-walk discharge.
• Reference performance characterization : after every 50 RW cycles, a pulsed-load characterization cycle plus EIS impedance sweep is performed. These reference cycles provide ground-truth capacity and internal-resistance estimates that anchor SOH labels.

File inventory

Files are distributed as MATLAB .mat archives (verified format). The full archive on NASA PCoE is named 11.+Randomized+Battery+Usage+Data+Set.zip and contains seven sub-archives (one per part), each holding several .mat files keyed by cell ID (RW1.mat, RW2.mat, …, RW28.mat).

A complete enumeration of files, sizes, and SHA-256 hashes will be written to the manifest YAML (bsebench-datasets/manifests/nasa_rw_2014.yaml) at upload time.

Known data-quality caveats (from prior literature using this dataset) :

• RW3 : temperature channel reportedly corrupted across most of the cell's life ; users should mask the thermistor column for this cell.
• RW20 : measurement vector reported as all-zero for most of the cell's life ; this cell should be excluded from estimator-quality metrics.

These caveats will be encoded as excluded_cells and channel_blacklist flags in the BSEBench Tier 2 harmonization manifest.

Why "raw mirror" tier

BSEBench follows a dual-tier dataset strategy :

• Tier 1 (this repo) — original .mat files, SHA-256 verified against the canonical NASA PCoE distribution. Used for provenance verification and audits.
• Tier 2 — see bsebench-org/nasa-rw-2014 for the harmonized BSEBench-canonical Parquet with consistent column names, BPX-1.1 sign convention, unified time origin, and cycle-number normalization across the seven parts.

For most filter benchmarking work, you want Tier 2. Tier 1 exists to let auditors verify that BSEBench's harmonization is faithful to the original.

Original source

Canonical NASA PCoE Data Repository entry :

https://www.nasa.gov/intelligent-systems-division/discovery-and-systems-health/pcoe/pcoe-data-set-repository/

Direct ZIP archive (NASA-hosted, S3 bucket of the PHM Society mirror) :

https://phm-datasets.s3.amazonaws.com/NASA/11.+Randomized+Battery+Usage+Data+Set.zip

Per BSEBench's Single Source of Truth policy, those URLs are recorded as citation metadata only. The HuggingFace mirror at bsebench-org/nasa-rw-2014-raw is the authoritative download endpoint for BSEBench reproducibility — independent of NASA's URL rotation, S3-bucket migration, or website restructuring (the canonical NASA URL has been observed to change at least twice in the past five years).

License

The NASA PCoE Data Repository distributes its datasets as U.S. Government Works. Per NASA's Open Data Policy, datasets generated under NASA sponsorship and not subject to a specific use-restriction notice are released into the public domain and are equivalent to Creative Commons Zero 1.0 Universal (CC0 1.0) for downstream licensing purposes.

The NASA PCoE Data Repository landing page itself states verbatim :

"Users employ the data at their own risk. Neither NASA nor the donators of the data sets assume any liability for the use of the data, or any system developed using the data."

"Publications making use of databases obtained from this repository are requested to acknowledge both the assistance received by using this repository and the donators of the data."

This Tier 1 mirror inherits the CC0 1.0 designation. Users are strongly encouraged to cite Bole, Kulkarni & Daigle (2014) per the NASA PCoE attribution request, even though attribution is not legally required under CC0.

For NASA's full Open Data Policy, see data.nasa.gov and NASA Earthdata Open-Data Policies.

Note for BSEBench users : the Tier 1 mirror is licensed CC0 1.0 matching NASA. The Tier 2 harmonized Parquet at bsebench-org/nasa-rw-2014 is also released CC-BY-4.0-compatible (BSEBench's own derived layer is CC-BY-4.0 ; underlying NASA data remains CC0). See Tier 2 card for details.

How to use

from huggingface_hub import snapshot_download

local_dir = snapshot_download(
    "bsebench-org/nasa-rw-2014-raw",
    repo_type="dataset",
)
# local_dir contains the original .mat files, SHA-256 verified

For the BSEBench-harmonized version with a consistent Python API and BPX-1.1 sign convention, prefer bsebench-org/nasa-rw-2014.

Citation

Cite the original NASA PCoE record (BibTeX above). BSEBench's contribution is hosting + harmonization, not the data itself.

If you also use BSEBench tooling, 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 schema lives at bsebench-org/bsebench-datasets/manifests/nasa_rw_2014.yaml.

Including : source.canonical_url, source.canonical_doi (none — NASA PCoE has no DOI assigned to this dataset as of 2026-04-29), every file's SHA-256 + size, the inherited public-domain / CC0 license, and the citation BibTeX above.