Early prediction of Li-ion cell failure from EIS derived from current–voltage time series

Abstract The ability to reliably detect the forthcoming failure of a rechargeable cell without removing it from its normal operating environment remains a significant goal in battery research. In this work we have cycled in the laboratory a previously-aged 3.2 A h, 3.6 V 18650 INR LiNi x Mn y Co <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mrow/> <mml:mrow> <mml:mn>1</mml:mn> <mml:mo>−</mml:mo> <mml:mi>x</mml:mi> <mml:mo>−</mml:mo> <mml:mi>y</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:math> O 2 cell for 300 d until failure was apparent, using a current waveform representative of use in an electric vehicle application. Electrochemical impedance spectroscopy (EIS) down to 5 µ Hz was also performed on the cell as a ‘gold-standard’ measure, at the beginning, end and part way through the cycling. Analysis of voltage and current time series data using both parametric (equivalent circuit model) and non-parametric (wavelet-based analysis) approaches allowed us to successfully reconstruct the EIS data. As the battery aged, impedance gradually increased at frequencies between 10 −4 Hz—10 −1 Hz. The increase accelerated around 50 d before the battery ultimately failed. The acceleration in rate of change of impedance was detectable while the cycle efficiency remained high, indicating that a user of the cell would be unlikely to detect any change in the cell based on its performance or by common measures of state-of-health. The results imply upcoming failure may be detectable from time series analysis weeks before any noticeable drop in cell performance.