PROGNOSTIC CLASSIFICATION PROBLEM IN BATTERY HEALTH MANAGEMENT

Batteries are crucial power sources for many engineering systems. Once the power source fails, the whole system completely stops performing its intended function. To avoid such catastrophic consequences, effective data‐driven prognostic tools that can accurately predict the failure in advance are highly desired. Effective failure prognosis plays an important role in maintenance decision‐making. Traditionally, most existing prognostic methods focus on the remaining useful life (RUL) prediction. Recently proposed prognostic methods have capability of real‐time online model updating. This chapter compares the logistic regression method and joint prognostic model (JPM), which is one of the most advanced data‐driven prognostic methods. It discusses a potential issue of having imbalanced data to provide important insights into the practitioners. The first category of methods typically models the time‐to‐failure data using parametric distributions such as Weibull distribution. The performance evaluation criteria used in the chapter are AUC, accuracy, and sensitivity.