Asymptotic Reduction Methods for the Parameter Estimation and Simulation of Lithium-Ion Batteries

Abstract

Physics-based battery models are a very powerful tool to study batteries as they provide insight on the internal states, such as the concentrations or potentials. However, these models are very complex and running them is computationally expensive, which makes them not suitable for many applications such as parameter estimation and battery control. Therefore, we need to reduce these models to others of lower complexity but that still retain most of the prediction capabilities.Asymptotic methods provide tools to reduce the models in a systematic way, keeping track of the validity and accuracy of the reduced model. These asymptotically reduced models present the advantage over ad hoc reductions that the reduced model is consistent with the full model and that we can check if the reduced model is valid in a given parameter set before even solving the model. This is especially useful in complex models with a lot of parameters, such as the Doyle-Fuller-Newman (DFN) model.Here we present an asymptotic model reduction for the particular applications of parameterisation and battery control. This reduction allows us to derive specific simple models with the complexity of the Single Particle Model (SPM) which, for a given application or operating conditions, capture the most relevant features of the full model. We first derive a reduced coupled thermal-electrochemical model and we parameterise it for a commercial cylindrical cell (LG M50). The complexity of this model makes it suitable for control applications. Then, we show how we can use asymptotically reduced models to determine some of the parameters of the model from experimental measurements of Galvanostatic Intermittent Titration Technique (GITT). This method ensures that the parameters obtained are consistent with the DFN model, and highlights the limitations and possible extensions of the technique.