A critical look at efficient parameter estimation methodologies of electrochemical models for Lithium-Ion cells

Abstract

Physics-Based Models (PBMs) offer a promising approach to develop advanced battery management systems that rely on information about the internal states of battery cells. The reliability of model predictions heavily depends on a proper parametrisation. However, the non-linear model structure, the high number of embedded parameters, and the experimental limitations, make the parametrisation procedure a difficult task. To tackle this issue, a myriad of approaches has been proposed in the research community, including physico-chemical characterisation techniques, non-invasive methodologies, or a combination of invasive- and non-invasive procedures, all aimed at maximising parameter identifiability. While a single solution may not exist, there is a recognised need to establish a systematic framework that can guarantee the correct estimation of model parameters. In this paper, we aim to review the key concepts and major challenges encountered in the field of parameter estimation of PBMs for the modelling of lithium-ion cells. Furthermore, the strengths and weaknesses of the current methodologies will be discussed based on previous attempts. Our analysis will lead to the conclusion that mixed methodologies, which combine invasive and non-invasive techniques, are promising approaches for a full-parametrisation of PBMs as they can maximise the identifiability of parameters. For the mixed methodology implementation, the essential steps that should be included are described: (1) parameter clustering, (2) design of optimal experiments, (3) sensitivity analysis, (4) selection of an optimisation algorithm for parameter fitting, and (5) the validation of the model. These steps must ensure, when possible, the convergence to a realistic parameter set and the model adaptability to multiple scenarios.