MagBridge-Battery dataset brings magnetic sensing to battery health diagnostics
6,760 synthetic magnetic signatures pair with degradation labels for AI-driven battery analysis
MagBridge-Battery v1.0 addresses a critical gap in battery diagnostics: while electrochemical signals from cell terminals dominate current methods, magnetic sensing can capture complementary information about internal state. However, development of magnetic-based diagnostics has been stymied by a lack of public datasets pairing magnetic measurements with degradation labels. The new synthetic dataset, released by researchers under a Creative Commons license, contains 6,760 magnetic-field signatures engineered to bridge real magnetic morphology from the Mohammadi-Jerschow Open Science Framework archive with state-of-health (SOH) labels from the PulseBat dataset.
The dataset comprises three subsets: 5,600 PulseBat-conditioned grounded samples, 600 synthetic sensor-anomaly samples derived from clean parents, and 560 low-voltage Regime-B extrapolation samples. A careful cell-disjoint split ensures zero leakage. Three benchmark tasks are defined: SOH regression (achieving R²≈0.77, which collapses to ~0 under label shuffle), second-life classification, and anomaly detection. This release provides a public benchmark for magnetic-sensing battery diagnostics while paired magnetic-electrochemical measurements remain scarce, enabling researchers to develop and compare AI models for non-invasive battery health assessment.
- Dataset contains 6,760 magnetic-field signatures with three subsets: 5,600 grounded, 600 anomaly, 560 extrapolation samples
- Defines three benchmark tasks: SOH regression (R²≈0.77), second-life classification, and anomaly detection
- Released under CC-BY-4.0 on Zenodo, with bridge code and benchmark suite under Apache-2.0 on GitHub
Why It Matters
Magnetic sensing could revolutionize battery diagnostics, but lacked training data; this synthetic bridge dataset unlocks AI model development for non-invasive health monitoring.