Concept for a Hybrid Performance- and Data-Based Model for State of Health Monitoring and Aging Prediction of Li-Ion Battery Packs

摘要

In the past, several aging models for li-ion batteries have been provided in literature. Many of them can be divided into either performance-based or data-based approaches. Both model domains have their specific strength and weaknesses, whereby most of them are complementary. To capture all relevant aging stress factors, development of performance-based models is very time-consuming and resource-intensive. Furthermore, they are parameterized on cell-level and do not display statistically induced aging scattering on system-level, wherefore their accuracy decreases during vehicle usage. The main drawback of data-based approaches is the necessity of a huge amount of data in order to display all relevant aging regimes in the training data and gain high accuracy. However, the high potential accuracy itself increases during continuous learning along the batteries lifetime. Still, if specific aging mechanisms occur towards the batteries end of life, which have not been observed in the training data, the black-box approach may fail to detect them. To overcome the deficiencies in state of the art aging modeling, the authors propose a new hybrid performance-and data-based approach for state of health monitoring and aging prediction of li-ion battery packs. The approach consists of several algorithms, which partly run onboard on a telematics unit and online on a server infrastructure. The concept is based on previous work by Baumann et al.. Onboard, the telematics unit receives data from the battery management system (BMS) and runs capacity and impedance estimation algorithms. Furthermore, the time-continuous data from the BMS needs to be aggregated to a reasonable data size, which can be transferred to the server. This is done by load spectrum analysis. Regarding the choice of the resolution of stress factor-classes, a trade-off between data-size and modeling accuracy has to be found. Therefore, the authors propose a variable class resolution based on areas with similar aging behavior, which are extracted by literature and own aging experiments. On the server side, an equivalent circuit model of the whole battery pack is regularly updated with the estimated parameters. The pack model considers parameter variations especially inside a parallel connection, which cannot be directly measured by the BMS and scales system behavior onto cell-level. Load spectrum analysis can then be performed on cell-level, before the aging relevant data is fed into the hybrid aging model. The hybrid aging model has three specific goals: 1) Prediction of the pack's available energy and power capability (in terms of capacity and resistance). This is done by interaction of an empirical aging model with a data-based regression algorithm. Prior estimated parameters serve as training data, whereby the probabilistic nature of the data is taken into account. 2) Detection of specific severe aging mechanisms, such as lithium plating, which will lead to fast battery degradation in the upcoming cycles. 3 ) Analysis of the pack's usage and occurring stress factors. This knowledge is needed as input to predict the further aging behavior based on the real usage profile. Hence, battery management functions could be adjusted in order to prolong the batteries lifetime.