README

This dataset consists of 123 industrial-grade large-format lithium-ion pouch cells released with the Nature article Discovery Learning predicts battery cycle life from minimal experiments. The article describes these cells as having diverse material-design combinations and cycling protocols. The source data are available from Zenodo and Code Ocean through the article's Data availability section.

The local processed directory contains 123 .pkl files, one file per cell. File names preserve the original Farasis cell id and source parquet path, for example:

Farasis_PB-b1_2_MP_029_single_parquet_with_soc_and_corrections_updated.pkl

If you use this dataset, please cite the original paper:

• Discovery Learning predicts battery cycle life from minimal experiments 2026 Nature

Source Data

The original dataset is available at:

• Nature article: https://www.nature.com/articles/s41586-025-09951-7
• Zenodo: https://doi.org/10.5281/zenodo.17654407
• Code Ocean: https://doi.org/10.24433/CO.5569470.v1

The Nature article states that the study data contain 123 large-format lithium-ion pouch cells. The Zenodo record contains these files:

• raw_data.zip
• rpt_data.zip
• number_relationship.xlsx
• readme.txt
• Parquet_analysis.py

The Zenodo readme.txt describes the source dataset as RPT .xlsx files and raw .parquet files for eight battery cell types. It states that the parquet files contain current, voltage, temperature, and other cycling information recorded at one-second resolution. It also states that number_relationship.xlsx maps 123 cells to 37 cell groups and records the cycle-life labels.

Local Raw Files

The local raw directory datasets/raw/Farasis/ contains:

The 14 .npy files are author-provided efc/soh arrays for seven MP or driving-protocol cells.

The local parquet files use columns including:

time_stamp
del_time
current (A)
voltage (V)
step_no
temp
source
rpt
soc
cv_hold
net_thrgh_ah
total_chg_thrgh_ah_rounded
total_dischg_thrgh_ah_rounded
total_energy_thrgh_kwh_rounded

Local Processed Files

The processed pickle files are generated from the local parquet files by process_scripts/preprocess_Farasis.py.

Each processed pickle contains BatteryML-style fields including:

cell_id
cycle_data
form_factor
anode_material
cathode_material
nominal_capacity_in_Ah
min_voltage_limit_in_V
max_voltage_limit_in_V
SOC_interval
charge_protocol
discharge_protocol

Each cycle_data entry contains fields including:

cycle_number
voltage_in_V
current_in_A
temperature_in_C
discharge_capacity_in_Ah
charge_capacity_in_Ah
time_in_s
attribute

Local File-Name Statistics

The 123 processed files are distributed across these filename-derived cell types:

The filename-derived protocol tokens are:

The seven MP or driving-protocol cells are marked as updated in the local filenames:

Farasis_PB-b1_2_MP_029_single_parquet_with_soc_and_corrections_updated.pkl
Farasis_PB-b2_1_MP_030_single_parquet_with_soc_and_corrections_updated.pkl
Farasis_PB-b2_1_MP_031_single_parquet_with_soc_and_corrections_updated.pkl
Farasis_PC-b1_2_MP_111_single_parquet_with_soc_and_corrections_updated.pkl
Farasis_PC-b1_2_MP_112_single_parquet_with_soc_and_corrections_updated.pkl
Farasis_PC-b2_2_Driving_041_single_parquet_with_soc_and_corrections_updated.pkl
Farasis_PC-b2_2_Driving_042_single_parquet_with_soc_and_corrections_updated.pkl

The Zenodo readme.txt states that these seven MP or driving-protocol cells have proprietary cycling portions removed from the raw data, leaving only RPT-related raw data.

Local Processed Metadata

The local preprocessor writes these metadata values:

These metadata values are taken from the local processed files and the local Farasis preprocessor. They should not be read as additional source-dataset claims beyond what is represented in this repository.

Labels

Label JSON files for this dataset are stored in datasets/processed/:

• Farasis_labels.json: author-provided labels transcribed from the source label table and keyed by processed Farasis file name in id order.
• repro_Farasis_labels.json: locally recalculated 90% SOH EFC labels based on the processed/raw Farasis data. The current local recalculation uses the cumulative discharge throughput column when available.

Notes

• This README does not claim operation-temperature ranges, electrolyte composition, or exact membership of the 37 source cell groups because those details are not fully represented in the local processed directory.
• Counts in the tables above were checked against the local files under datasets/raw/Farasis/ and datasets/processed/Farasis/.