Optimal Experimental Design for Parameterization of an Electrochemical Lithium-Ion Battery Model

We consider the problem of optimally designing an excitation input for parameter identification of an electrochemical Li-ion battery model. The conventional approach to performing parameter identification uses standard test cycles. In contrast, we optimally design the input trajectory to maximize parameter identifiability in the sense of Fisher information. Specifically, we derive sensitivity equations for the electrochemical model. This approach enables parameter sensitivity analysis and optimal parameter fitting via gradient-based algorithms. This paper presents a general systematic approach to identify the electrochemical parameters in a non-invasive way. First, we group parameters into two sets: (i) equilibrium parameters, and (ii) dynamical parameters. We also divide the dynamical parameters into subsets by calculating orthogonalized sensitivity, which mitigates linear dependence between parameters. A large number of input profiles have been devised to constitute an input library. Then, the optimal inputs are selected from the input library to maximize the Fisher information, via convex programming. Using this framework a number of relevant experiments are obtained to parameterize. To validate our approach experimentally, we consider a 18650 Lithium nickel cobalt aluminum oxide battery. Compared to the conventional approach, our proposal achieves lower voltage RMSE across all experimental testing cycles.