Battery state observation and condition monitoring using online minimization

In this paper, the performance of a particular state observer algorithm is investigated for concurrently estimating the state of charge of a Li-Ion battery and the deviation of characteristic parameters from their initial values in order to diagnose the aging behavior of the system. The estimation is based on an online minimization approach that has been chosen in particular due to the significant nonlinear behavior of the battery dynamics. The method has been introduced as the “Newtonstep observer” but could also been interpreted as a “sensitivity-based estimation”. The analysis is first made by simulations for comparison with known reference values. Here, we also compare the performance of Newton-type, Gauss-Newton-type and gradient-based optimizations. In a second step, experimental data from a battery that has been subject to an aging process with 400 charging/discharging cycles is used. Despite the fact that for the battery actual reference parameters are unavailable, the results show physically plausible trends. Moreover, the estimated capacity fading is particularly in very good agreement with the data provided by the manufacturer.