The functionality and reliability of Li-ion battery as major energy storage device has received more and more attentions from a wide spectrum of stakeholders including federal/state policymakers, business leaders, technical researchers, environmental groups and general public. Failures of Li-ion battery not only result in serious inconvenience and enormous costs, but also increase the risk of inducing catastrophic consequences. In order to prevent severe failure from happening and optimize the Li-ion battery maintenance schedules, breakthroughs in prognostics and health monitoring of Li-ion battery with emphasis on fault detection and correction, remaining-useful-life prediction and assist maintenance scheduling must be achieved.
This paper firstly reviews recent research and development of health monitoring and prognostics for a variety of battery chemistries and summarizes the techniques, algorithms and models used for state-of-charge (SOC) estimation, current/voltage estimation, capacity estimation and remaining-useful-life (RUL) prediction. Based on the understanding of these researches, two data-driven approaches have been proposed to evaluate and predict the health status of Li-ion battery. The first approach, Gaussian mixture modeling, is an unsupervised learning method which doesn’t require capacity as target. This is surely a pleasant feature since the capacity, though a good indicator of battery health, is neither easy nor cheap to acquire. Instead, the underlying pattern of input space is studied and the useful information (CV) extracted from it discloses the health status of Li-ion battery. The other approach, adaptive regressions, is a supervised learning method. Though targets are required this time, capacity is still not used. Instead, an important feature is found to have close relationship with capacity while easy to get in real world application. Based on this relevant feature, an adaptive framework for health assessment and prognostics is introduced. Results show that the adaptive regression approach is exceptionally good in making near-term predictions. The long-term prediction, though not satisfactory in terms of accuracy, well addresses the so-called “psychological perception” of human being. Therefore, the users of electric cars are protected against unreasonably excessive panic about battery breakdown. The otherwise maintenance and/or replacement costs could also be saved.
The Li-ion battery datasets from NASA Ames Research Center are used to develop models for health assessment and prediction of Li-ion battery. The run-to-failure test consists of three operational profiles: charging, discharging and impedance tests. Evidence has shown that the impedance tests are not only costly and time-consuming to implement, but also impractical in real-world tasks. Moreover, impedance tests permanently damage battery health by requiring extra charge and discharge cycles and result in difficulty in modeling degradation trends of Li-ion batteries. Therefore, it is strongly recommended that impedance tests should be removed in the future run-to-failure tests on Li-ion batteries.