Searching for a Pulse: Evaluating the Use of Rapid DC Pulses for Diagnosing Battery Health, State-of-Charge, and Safety
Paul Gasper, Nina Prakash, Bryce Knutson, Thomas Bethel, Katrina Ramirez-Meyers, Amariah Condon, Peter M. Attia, Matthew Keyser
Abstract
Rapid electrochemical diagnostics, like DC pulse sequences or electrochemical impedance spectroscopy, are known to be useful for capacity prediction. However, it is unclear how previous results will map to different cell types and use cases and whether rapid diagnostics are useful for remaining useful life prediction or for detecting potential safety issues. To that end, we have collected a data set with ∼50,000 DC pulse measurements from four types of commercial lithium-ion batteries to enable training of state-of-charge, health, and safety diagnostic models via machine-learning. We demonstrate that 120-second DC pulse sequences can be used to predict capacity with 2%–9% average error, which can separate high- from low-capacity cells with only a 0.3% false positive rate but is not accurate enough to estimate remaining useful life. We also find that no safety related targets can be accurately predicted, highlighting the critical need for other non-invasive methods to diagnose battery safety.