Rapid diagnosis of thermal and electrochemical defect and aging mechanisms of lithium-ion batteries

The pervasive application of lithium-ion batteries has significantly reduced costs and rapidly increased global cell production. While this supports national and international targets for decarbonization and electrification, particularly of the transport sector, it also poses substantial challenges. Traction batteries from damaged or retired electric vehicles demand efficient approaches to classify their suitability for second-life applications, assess recyclability, and identify safety concerns. In second-life scenarios, rapid diagnosis and decision-making are essential for economically feasible battery handling. Existing research on battery diagnostics has largely focused on accurately predicting aging or defects in well-referenced cells. Universal approaches for analyzing cells with unknown chemistries, states of charge, and states of health remain largely unexplored, despite their potential to significantly increase re-use and recycling rates. This work investigates different approaches for the rapid diagnosis of lithium-ion batteries and their combination to assess second-life and recycling viability. Data-driven identification of cell chemistries, particularly distinguishing lithium nickel manganese cobalt oxide from lithium iron phosphate, based on partial open circuit voltage curves is investigated. Synthetic data was generated to improve generalizability across different states of charge and health. The trade-off between measurement speed and classification accuracy is evaluated, providing insights into the influence of measurement steps and charged or discharged capacity. State of health estimation represents another crucial aspect of usability assessment. Experimental campaigns inducing specific defect and aging mechanisms, such as lithium plating and solid electrolyte interphase growth, were conducted alongside electrochemical impedance spectroscopy reference measurements. The resulting data were combined with publicly available impedance data from various cells to investigate rapid state of health determination for unknown lithium-ion batteries using neural networks. The most relevant frequency range was identified between 1 kHz and 100mHz, demonstrating that multisine impedance measurements as short as four seconds can predict state of health with a mean absolute error of 1.6 %. Finally, a novel method for assessing the state of usability of end-of-life cells is introduced, enabling the combination of different diagnostic methods and tools.