Estimating and Identifying Parameters from Charge-Discharge Curves of Lithium-Ion Batteries

Abstract

Lithium-ion battery, for its promising applications in energy storage systems and electric vehicles, has been increasingly popular since its commercialization. Due to the needs to monitor, predict and control the status of lithium-ion batteries, physical model based management systems have been intensively studied and developed by many researchers(1). The accuracy and predictability of the model used are of great importance to these systems, it is essential to get the parameters needed in the battery models. Estimating parameters for the lithium-ion batteries is challenging due to the complexity of the governing equations, and the possibility of multiple set of parameters that might provide the same accuracy of fitting discharge curves. Parameter estimation of various lithium-ion battery systems has been done for different models, including equivalent circuit model(2), single particle model(3) and pseudo 2D (P2D) model(4, 5). Most of these models are built on known open circuit voltage curves for individual cells, base parameters for cathode/anode thickness, porosities etc. P2D model, also referred to as Doyle-Fuller-Newman (DFN) model, is a first principal model that has been well studied and experimentally validated. Previous attempts to reduce the computational cost of parameter estimation by incorporating the reformulated P2D model to parameter estimation has been done by V. Ramadesigan et al (6). Based on control theory, studies on model and parameter identifiability has been conducted by researchers, which typically concludes that only certain number of parameters can be estimated with good confidence and also identifies operating conditions in which certain parameters will be more sensitive for estimation(7-9). In this presentation, we will show how physical intuition can help converge on some of the parameters when standard estimation methods fail. We will attempt to address the possibility and relative importance/impact of estimating all the parameters needed for the DFN model from charge-discharge curves. Special attention will also be paid to the computation time to enable online state and parameter estimation. This will be facilitated using our past results in model reformulation(10). We will also demonstrate how minor modifications with transients can help identify and converge on parameters when standard estimation methods fail. We will present how perturbation methods can help to increase the sensitivity of certain parameters without compromising on charge stored or life of the battery. It is observed that estimation of parameters can be improved when the right range of frequencies and amplitudes are used in the transient signals provided. This includes sinusoidal input typically provided in NLEIS(11). Acknowledgements The authors are thankful for the financial support of this work by the Clean Energy Institute (CEI) at the University of Washington, Washington Research Foundation (WRF), and the U.S. Department of Energy’s Advanced Research Projects Agency-Energy (ARPA-E).