Impedance-based forecasting of lithium-ion battery performance amid uneven usage

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Accurate lithium-ion battery performance forecasting is critical for electric vehicle (EV) safety and reliability.
• Existing battery health prognostics often lack real-world applicability due to uniform usage patterns in lab settings.
• Practical EV battery operation involves significant variability in usage, complicating performance prediction.

Purpose of the Study:

• To develop a robust method for predicting future lithium-ion battery discharge capacity.
• To address the challenge of forecasting battery performance under variable and unknown usage histories.
• To provide calibrated uncertainties alongside performance predictions.

Main Methods:

• Utilized a dataset of 88 commercial lithium-ion coin cells.
• Employed a combination of electrochemical impedance spectroscopy (EIS) measurements and probabilistic machine learning algorithms.
• Tested predictions against multistage charging and discharging with randomly varied currents between cycles.

Main Results:

• Future discharge capacities were predicted with calibrated uncertainties.
• The model accurately forecasted performance using a single EIS measurement before charging, irrespective of usage history.
• Results demonstrated robustness across different cell manufacturers, cycling protocols, and temperatures.

Conclusions:

• Electrochemical impedance spectroscopy combined with probabilistic machine learning offers a powerful tool for battery health prognostics.
• Battery health can be more effectively characterized by a multidimensional vector than a single scalar value.
• This approach enhances the reliability and safety of electric vehicles by enabling accurate battery performance forecasting.