Statistical methodology for predicting the life of lithium-ion cells via accelerated degradation testing
Statistical models based on data from accelerated aging experiments are used to predict cell life. In this article, we discuss a methodology for estimating the mean cell life with uncertainty bounds that uses both a degradation model (reflecting average cell performance) and an error model (reflecting the measured cell-to-cell variability in performance). Specific forms for the degradation and error models are presented and illustrated with experimental data that were acquired from calendar-life testing of high-power lithium-ion cells as part of the U.S. Department of Energy's (DOEs) Advanced Technology Development program. Monte Carlo simulations, based on the developed models, are used to assess lack-of-fit and develop uncertainty limits for the average cell life. In addition, we discuss the issue of assessing the applicability of degradation models (based on data acquired from cells aged under static conditions) to the degradation of cells aged under more realistic dynamic conditions (e.g., varying temperature).