Estimating the state-of-charge SoC and state-of-health SoH of batteries is a key task for battery management systems. Observer based approaches such as the Dual-Extended-Kalman-Filter for joint SoC- and SoH-estimation have been proposed more than two decades ago and countless publications since then have shown the applicability of similar and advanced estimation methods.
Two questions essential for the practical implementation of these observers however require more attention and shall be the foci of this presentation. Firstly, stability and robustness are difficult to prove if the system is non-linear, complex and incorporates several values to be estimated simultaneously. How the estimation of the parameters of the underlying battery model – which are the basis for the SoH – and the SoC interact and how many states and parameters can be estimated simultaneously is poorly understood. Secondly, the performance of the observers strongly depends on the tuning of their hyperparameters; strict design procedures only exist for idealized systems.
These issues and possible solution approaches shall be addressed in the presentation from two directions. Based on a strongly reduced battery model, the simultaneous observability of SoC, internal resistance and battery capacity is discussed analytically, which yields interesting results under which practically relevant load situations the system is observable or not (fig. 1).
Implications on load-dependent observer control will be discussed.
For more complex and non-linear observer structures a modular process and tool chain for multi-objective optimization of battery observers via hyper-space-exploration will be presented (fig. 2). Within a virtual experiment environment, variations in use-case and design-space can be analyzed and hyperparameters are optimized for multiple target indicators such as accuracy of the estimation and convergence speed. Both sensitivity and trade-off analyses provide valuable insight into interdependencies and provide a systematic approach on observer tuning.