Statistics of Battery Failure
Steve Harris
Lawrence Berkeley Lab
The battery community, implicitly or explicitly, usually takes failure to be deterministic rather than statistical. This means that there should be a function (model) which predicts a single value for life as a function of critical parameters such as C-rate, temperature, electrode properties, etc. Variability is rarely explicitly discussed, but when it is, it is further assumed that any variability in life is Gaussian and that with sufficiently close control of the appropriate parameters, the variability can be always be reduced. In this talk I will make the case that battery failure is intrinsically statistical, which means that there is no single function (model) that predicts the lifetime of a battery, just as there is no model that can predict lifetime of a person or a gear. On the other hand, as I will show with standard statistical methods, it is possible to predict the time-evolution of the distribution of lifetimes with considerable accuracy using capacity vs cycle data. Furthermore, the analysis offers a way to qualitatively classify the type of degradation that the cells are undergoing.
