Electrode-Specific Degradation Diagnostics for Lithium-Ion Batteries with Practical Considerations

Abstract

Li-ion batteries inevitably degrade with cyclic usage and storage time. Close to end-of-life batteries can no longer meet their performance requirements and the likelihood of occurring catastrophic failures increases. Thus, an accurate diagnosis of their state of health over long-term use has become a critical function for reliable and safe battery management systems, especially for vehicle electrification and large scale energy storage systems. Degradation of batteries is typically quantified at the cell level with capacity loss and power fade, however different usage conditions and environmental factors can contribute to the degradation of batteries differently. Therefore, typical cell-level lumped degradation metrics are not sufficient to give a full explanation of battery state of health.

This dissertation presents approaches for the diagnosis of electrode-specific degradation of Li-ion batteries considering a variety of practical aspects in real-world applications such as the half-cell potential change, the partial data availability, data acquisition method, and practical charging rate. The electrode-specific degradation diagnosis is performed by model-based identification of the individual electrode state-of-health (eSOH) parameters, electrode capacity and utilization range.

The advancements contributed by this dissertation are summarized as follows. First, a novel diagnostic algorithm is proposed by combining the terminal voltage fitting process with the peak alignment method to improve electrode parameter estimation confidence. The proposed method addresses the half-cell potential change of the positive electrode due to the chemical aging of the metal oxide. The diagnostic result is experimentally verified with large-format prismatic commercial cells. The second practical consideration is partial data availability. In practice, the full range of OCV measurement is not obtainable without the designated offline diagnostic test. With the limited data, the accuracy of parameter estimation becomes questionable. Therefore, the achievable estimation error bound is analyzed with respect to partial data windows through the Cramer-Rao Bound and confidence interval. The result shows that the eSOH estimation improves when a data window includes slope changes of electrode half-cell potential. This fundamental limitation is applied in data-driven approach to provide data-requirements for machine learning of battery cycle life prediction. Third, continued from the partial window idea, a time-optimal current profile is proposed to enable direct measurement of pseudo-OCV data for the desired range without a long relaxation period. By allowing bi-directional charging, the proposed time-optimal control problem identifies a proper sequence of charge/discharge pulses and successfully reduces total data acquisition time by more than 60% in both simulation and experiment, showing a possible way to implement the developed OCV-based electrode degradation diagnostic algorithm. Fourth, the feasibility of the electrode-specific degradation diagnostics is studied for real-world charging conditions where the typical charging current rate is usually higher (e.g. C/5) than C/20 of pseudo-OCV data. With increasing charging rates, the individual electrode's electrochemical features is obscured, and the overpotential due to internal resistance needs to be estimated concurrently, making the eSOH estimation challenging. An adaptive algorithm with a data selection strategy is proposed to deal with the estimation of both resistance and electrode SOH parameters. Lastly, the potential of the physics-guided machine learning approaches is explored with two case studies for Li-ion battery degradation diagnostics and prognostics.