Machine learning for optimised and clean Li-ion battery manufacturing: Revealing the dependency between electrode and cell characteristics

The large number of parameters involved in each step of Li-ion electrode manufacturing process as well as the complex electrochemical interactions in those affect the properties of the final product. Optimization of the manufacturing process, although very challenging, is critical for reducing the production time, cost, and carbon footprint. Data-driven models offer a solution for manufacturing optimization problems and underpin future aspirations for manufacturing volumes. This study combines machine-learning approaches with the experimental data to build data-driven models for predicting final battery performance. The models capture the interdependencies between the key parameters of electrode manufacturing, its structural features, and the electrical performance characteristics of the associated Li-ion cells. The methodology here is based on a set of designed experiments conducted in a controlled environment, altering electrode coating control parameters of comma bar gap, line speed and coating ratio, obtaining the electrode structural properties of active material mass loading, thickness, and porosity, extracting the manufactured half-cell characteristics at various cycling conditions, and finally building models for interconnectivity studies and predictions. Investigating and quantifying performance predictability through a systems' view of the manufacturing process is the main novelty of this paper. Comparisons between different machine-learning models, analysis of models’ performance with a limited number of inputs, analysis of robustness to measurement noise and data-size are other contributions of this study. The results suggest that, given manufacturing parameters, the coated electrode properties and cell characteristics can be predicted with about 5% and 3% errors respectively. The presented concepts are believed to link the manufacturing at lab-scale to the pilot-line scale and support smart, optimised, and clean production of electrodes for high-quality Li-ion batteries. [Display omitted] •Combining battery manufacturing experiments and machine learning models.•Predicting electrode structural properties based on manufacturing parameters.•Predicting the battery cell performance via electrode properties.•Robustness analysis of prediction models to data size and measurement noise.•Structural and electrochemical characteristics predicted with less than 5% and 3% error.