Distinguish an Outlier Cell from Genuine Aleatoric Variability

Determine whether the unusually large capacity-loss spread observed for the cell tested at 25°C and 100% state of charge represents a true outlier or genuine aleatoric variability in battery aging under that condition, using a larger set of replicate cells.

Background

The field prediction uncertainty is strongly influenced by the laboratory calendar-aging test point at 25°C and 100% state of charge, where one of the three tested cells exhibits substantially greater capacity loss than the other two. Replacing this cell’s measurements with synthetic data having a smaller spread reduces the predicted end-of-life uncertainty by approximately 50%.

Because only three cells were tested at the relevant condition, the available data cannot establish whether the anomalous cell is an atypical outlier or whether the observed dispersion reflects genuine stochastic variability in the degradation process. Additional replicate cells are needed to characterize the underlying variability reliably and to support risk-informed decisions about whether such observations should be retained or excluded from model training.