Discovery Benchmark

Discovery Benchmark

A reference dataset comprising ~250 lithium-ion coin cells systematically fabricated via automated assembly equipment at DIGIBAT facility, Imperial College London, UK. Designed to span common cathode/anode/electrolyte chemistries, as a shared benchmark for battery-informatics tooling.

Contents

The dataset is provided as an RO-Crate conforming to the ELN File Format standard. A single .eln file export contains subdirectories per item with its attached raw data and metadata as recorded in datalab, as well as top-level metadata available in the ro-crate-metadata.json. The .eln file is a zip file; users on operating systems that rely on file extensions may need to append .zip to the filename to coerce it.

A second version of the .eln file is included, which has been annotated with BattINFO and EMMO ontology terms in the ro-crate-metadata.json. The added fields explicitly define the relationships between the cells and other elements in the dataset, making it much easier to reuse the dataset in the future.

The original Excel recording of the dataset is also provided in ./Discovery-Benchmark.xlsx.

Characterisation

Precursor-level characterisation: Microstructural, morphological, and chemical properties—including SEM/TEM, XRD, and XPS are mapped directly to individual electrode precursor records (e.g., P025-ELD-001) rather than individual cells.

Cell-level characterisation: In-situ and operational testing data—such as Neware cycling exports (.xlsx) and BioLogic CV/EIS data (.mpr)—are anchored directly to the respective coin cell entry (e.g., P025-CEL-001).

Visual records: Raw cell assembly photos are attached directly to each cell record. (Note: electrode alignment images are only available for cell batches built using the EQP-D02-Cellerate).

Mass-correction Note

The original charge/discharge capacities recorded in cell-side files used the total electrode disc mass (e.g. 23 mg) rather than the active material mass after subtracting current collector foil (e.g. 20 mg). Capacities have been recalculated; the corrected nominal capacity is shown on each cell record as characteristic_mass / nominal_voltage. Some cells turn out to be limited on the carbon side after recalculation — for those the nominal capacity is taken as min(cathode, anode). The per-file raw capacity values in the cycling exports are therefore stale; the cell-record value is the source of truth.

Infrastructure & Funding

Automated energy research facility: DIGIBAT, Department of Chemical Engineering, Imperial College London, funded by EPSRC & UKRI strategic equipment grant (EP/W036517/1).

Ingestion Pipeline: The ingestion code was developed by datalab industries ltd. and is accessible at https://github.com/datalab-industries/digibat-datalab-deployment.

Sponsorship: This campaign to generate open, multimodal battery datasets for the research community is supported by the Henry Royce Institute, Department for Science, Innovation & Technology, and UK Research and Innovation (UKRI2711).

Project Contributions & Framework

A collaborative milestone involving the multi-disciplinary deployment of laboratory automation, data engineering, and informatics tools:

Data Creators: Dr Robert Hunter, Dr Niamh Hartley

Data Curators: Dr Matthew Evans (datalab) Dr Lucas Garcia Verga (Data architecture), Dr Felix Mildner (BattINFO annotation, DigiBOT creator), Dr Ben Smith (datalab), Shirley Xiong (CellSeer visualisation)

Project Manager: Dr Jingyu Feng

Project Leaders: Prof Magda Titirici, Prof Aron Walsh, Dr Samuel J Cooper