Data-driven prognostics for Lithium-ion battery health monitoring

Li-ion batteries are a popular choice of rechargeable battery for use in many applications like portable electronics, automobiles as well as stationary applications for providing uninterruptable power supply. State of Charge (SoC) and State of Health (SoH) are important metrics of a Li-ion battery that can help in both battery prognostics and diagnostics for ensuring high reliability and prolonged lifetime. The ML algorithms available in the literature for SoC and SoH prediction involves use of various derived features rather than directly measurable features making it difficult for industrial applications. In this work, we use battery data obtained from different batteries to develop supervised models that can be used for the on-line estimation of SoC and SoH. This work involves two parts: a) developing a classifier based on SoH b) dynamic prediction of battery SoC given the past operational data of current, voltage, and temperature of the battery which are easily measurable. Random forest algorithm is used for battery site classification based on the SoH data available from the manufacturer. The battery SoC estimation is performed using a random forest algorithm and Neural network-based NARX model.