Machine Learning Approaches for State of Health Estimation in Lithium-Ion Batteries: An Exploratory Analysis of the NASA Battery Dataset

Abstract

Estimating state of health (SOH) is crucial to the safe and effective use of lithium-ion batteries in electric vehicles and home battery storage. This paper presents a focused review of state of health (SOH) literature, as well as exploratory data analysis (EDA) of the NASA lithium-ion battery aging data set. Our review shows a trend away from measurement and physics-based models towards data-driven machine learning and deep learning models, which make use of voltage, current, temperature, capacity, impedance and cycle data. The NASA dataset has 7,565 entries for 34 batteries and includes charge, discharge and impedance cycles. Our EDA shows a significant variation in the number of cycles for different cells, discharge capacity fade for a group of selected batteries, and degradation behavior in both the charge and discharge cycles. In battery B0047, discharge capacity drops from 1.674 Ah to 1.449 Ah during the recorded cycles and the SOH curve exhibits a downward trend. The early and late discharge profiles also have sagging voltage. The results indicate that capacity, resistance and dynamic behavior are good indicators for SOH calculation in practical applications.