Machine Learning Workflow to Track Degradation Mechanisms in Industrial Lithium Ion Battery Cells

The degradation of lithium-ion battery (LiB) cells is a complex process resulting from the interaction of multiple phenomena. Understanding LiB cell aging requires tracking measurable effects of different degradation mechanisms, known as degradation modes. These include the loss of active material in the electrodes, loss of lithium inventory, and conductivity loss. The aging mechanisms are directly correlated with electrochemical cycling protocols and operation conditions of the LiB cell, such as parameters related to the type and rate of current in the charge and discharge, voltage cutoff conditions, temperature, break periods, state of charge, and cycle number.<br/><br/>Using machine learning techniques to uncover the interrelations between aging conditions/protocols and mechanisms can trigger tremendous progress in designing battery management systems capable of extending cell lifetime in Electric Vehicle (EV) applications [1]. In this work, we introduce a novel workflow that combines various machine learning techniques with electrochemical LiB cell state of health diagnostic analysis, facilitating the discovery of the impact of a broad set of cycling conditions/protocol parameters on aging modes [2]. This approach allows for deriving the possible aging mechanisms that might manifest in each distinct condition.<br/><br/>While numerous studies have employed machine learning techniques on LiB cell datasets, they often focus on a limited number of cycling stressors (e.g., C-rate for charge and discharge), and the datasets typically originate from tests conducted in academic laboratories. In this study, a comprehensive industrial dataset, encompassing a broad range of cycling conditions and protocols to simulate real-world usage of LiB cells in Electric Vehicles (EVs), is utilized. We explore how more than 10 stressors impact the aging modes of cells and identify the predominant degradation mechanisms based on cycling conditions. The analytical capabilities of our machine learning tool are demonstrated through various application examples, and strategies for mitigating aging in automotive applications are discussed. Finally, we consider how our machine learning workflow holds the potential to propel current battery management systems in EVs to the next level.<br/><br/>[1] Artificial Intelligence Applied to Battery Research: Hype or Reality, Battery Research: Hype or Reality, Teo Lombardo, Marc Duquesnoy, Hassna El-Bouysidy, Fabian Årén, Alfonso Gallo-Bueno, Peter Bjørn Jørgensen, Arghya Bhowmik, Arnaud Demortière, Elixabete Ayerbe, Francisco Alcaide, Marine Reynaud, Javier Carrasco, Alexis Grimaud, Chao Zhang, Tejs Vegge, Patrik Johansson, and Alejandro A. Franco*, Chem. Rev. 2022, 122, 12, 10899–10969, 2021.<br/><br/>[2] Amina El Malki, Mohamed Ati, Mark Asch, Alejandro A. Franco, to be submitted (2023).