Lifetime prediction of lithium-ion battery (LIB) cells has been widely contemplated by means of various methodologies from physics-based aging models with good interpretability in aging mechanisms to machine learning frameworks with high precision in extrapolation capabilities. Furthermore, forecasting the lifetime in a quick and accurate manner plays an essential role in the performance analysis of LIBs, being for example essential for deciding optimal warranty periods and quality control designs. However, there is a trade-off between an early prediction and a prediction of high accuracy.
This study discusses the statistical concept of censoring in reliability and its role in lifetime testing and the optimal design of life testing experiments, targeting at experiments of shorter duration while controlling the precision of the predictions. We focus on Type-II censoring, under which the experiment is terminated when a pre-specified number of failures is observed. Maximum likelihood estimation and confidence intervals for expected lifetimes of LIB cells based on complete and censored data sets are compared for real and simulated data. We show how planning an experiment under censoring leads to experiments of shorter duration without sacrificing precision, discussing the optimal specification of the termination condition as well. Utilizing reliability concepts, statistical methods enable lifetime testing procedures of sufficient confidence in prediction for researchers and manufacturers.