MagBridge-Battery: A Synthetic Bridge Dataset for Li-ion Magnetometry and State-of-Health Diagnostics

MagBridge-Battery v1.0 is a synthetic dataset of 6,760 magnetic-field signatures for lithium iron phosphate (LFP) cells, produced by a bridging procedure that conditions the Mohammadi-Jerschow OSF magnetometry archive on electrochemical labels from the PulseBat dataset. The release contains 5,600 PulseBat-conditioned grounded samples, 600 synthetic sensor-anomaly samples derived from clean parents (four subtypes: sensor_dropout, calibration_drift, temporal_warp, periodic_interference), and 560 low-voltage Regime-B extrapolation samples.

A cell-disjoint, parent-child-leakage-free primary benchmark split is verified to contain zero overlapping cells, zero cross-split parent-child pairs, and zero sample-ID overlap. Four benchmark tasks are defined: SOH regression, second-life classification (cutoff SOH = 0.85), three-class anomaly detection, and four-class anomaly subtype classification.

Bridge validation includes structural sanity invariants, distributional KS tests at grounded anchors, and a controlled label-shuffle ablation that collapses SOH regression from R² ≈ 0.77 to R² ≈ 0, confirming that the bridge encodes input SOH non-trivially rather than producing label-aligned artifacts.

Users are kindly requested to cite both this dataset DOI and the associated paper (see CITING.md in the bundle). Code, paper source, and reference implementations are available on GitHub at https://github.com/SakthiGs/MagBridge-Battery.

ARXIV: https://arxiv.org/abs/2605.20240