Linking Failure Mode Analysis and Machine Learning to Accelerate Life Prediction

Batteries continue to see rapid advancement both in materials selection and in emerging use cases. In all cases as new options for battery use and design arise there is a need to validate performance and to understand failure modes. Traditionally validation processes would require years of testing and immense use of resources to reach acceptable certainty that a new battery would function in an intended use case. Recent emergence of the use of combining failure mode analysis with machine learning and other advanced data analytics provides the opportunity to dramatically reduce the time needed for validation and to more quickly bring new discoveries and uses cases from the bench top to the consumer.<br/><br/>In this discussion several different approaches to reducing the time needed for perfromance validation will be discussed. This includes both the importance of appropriate and targeted data collection as well as the data analytic methods that can support rapid validation. The key to both the advanced analytics and data collection is the need to ensure a link to physical processes which aids in battery failure mode identification. The link also allows extended data sets to be generated using different modeling or synthetic data approaches to be used for enhanced training purposes.<br/><br/>Using a combination of experimental and synthetic data it will be shown that enhanced predictions are possible and that the time and resources needed to make predictions can be dramatically reduced. As an example set different predictions for graphite/NMC batteries used in standard and fast charge conditions will be highlighted.