State Estimation of Lithium-ion Batteries
Time: Wed 2021-06-09 14.00
Location: https://kth-se.zoom.us/j/63931176711, Stockholm (English)
Subject area: Chemical Engineering
Doctoral student: Xiaolei Bian , Kemiteknik
Opponent: Associate Professor Huazhen Fang, University of Kansas
Supervisor: Associate Professor Longcheng Liu, Kemiteknik; Professor Jinying Yan, Kemiteknik, Vattenfall AB, R&D
Abstract
To guarantee the safety operation, the key states of lithium-ion battery, e.g., the state of charge and the state of health, must be estimated and monitored accurately. This thesis is mainly to develop models and algorithms to accurately and robustly estimate the key battery states, based on the available measurements i.e., current and voltage. All the work is based on four published papers and can be divided into three parts.
The first part of this work presents a two-step parameter optimization method for online state of charge estimation of lithium-ion battery. The particle swarm optimization is exploited for model parametrization and extended Kalman filter tuning. Within this particle swarm optimization-based framework, the searching boundary is derived by scrutinizing the error transition property of the test system, which can narrow the searching region and increase the computational efficiency. In general, the proposed method can well exploit the potential of model-based estimators, leading to a robust model compatibility and optimized performance.
In the second part of this thesis, two novel models are developed to estimate the state of health of lithium-ion battery. The first one is an open circuit voltage-based model, which describes the open circuit voltage as a function of the state of charge by a polynomial, with a lumped thermal model to account for the effect of temperature. It requires a prior learning from the initial constant-current profile. The second model is an incremental capacity analysis-based model, which defines the dependence of the state of charge on the open circuit voltage using a capacity model. It can be learning-free, with the parameters subject to certain constraints. Both models use an equivalent circuit model to characterize the constant-current profiles and a nonlinear least squares method to identify the involving parameters. These two models are validated by aging experiments, and the results show that both can give accurate state-of-health estimation.
The third part of the thesis introduces a fusion-type state-of-health estimator by combining the model-based profile reconstruction and the incremental capacity analysis-based state estimation. The above-mentioned open circuit voltage-based model is employed here to mitigate the noise-induced unfavorable numerical conditions and to modify the incremental capacity curves. Leveraging the modified incremental capacity curves, a set of feature-of-interests are extracted and evaluated, and several cautiously selected ones are used to estimate the state of health of lithium-ion battery. Long-term cycling tests on different lithium-ion batteries are used for validation. This fusion-type method has comparable accuracy and better robustness, compared with the model-based methods. Moreover, the proposed estimator has a good generality to different batteries and also promises an excellent robustness against cell inconsistency, noise corruption, temperature variety, and profile partialness.