GRaylevel Analysis of Particles in ElectrodeS (GRAPES); Data Workflows and Analytics for Computed Tomography (CT) Data of Degradation in Battery Materials

GRAPES is a python module designed to help researchers analyze electrode particles in tomography data. GRAPES transforms 3D image data into tabular dataframes that facilitate the rapid and intuitive processing of particles and particle properties relevant to the study of battery materials. Each row in the dataframe represents a unique particle with 20+ columns of particle characteristics such as sphericity, surface area, and volume. Additionally, battery degradation specific properties are included such as damage factor, and radial layer damage factors. The data transformation from image volumes to GRAPES dataframe can be executed in one line and allows users to more rapidly understand the relationship between different particle characteristics in their datasets. Furthermore, this format of large tabular data can be easily integrated with open-source machine learning libraries in python (sklearn, pytorch, etc.) to train classical machine learning models such as multiple linear regression, random forests, and gradient boosting.<br/><br/>In this work we report on two use case examples with a specific focus on the underlying data workflow used. These case examples are the degradation of silicon anode particles and NMC cathode particles during cycling. We cover the image processing, image registration, and image segmentation (binary and instance segmentation) required to prepare a time-lapse CT dataset. Additionally, we will discuss the importance of removing edge cases from the GRAPES dataset. This allows us to use GRAPES and other analytics to track how particles are damaged at different cycles, the spatial distribution of damaged particles, the regions of highest damage within particles, and track the percentage of particles that observe particular behaviors during cycling. Through this work we have demonstrated a state-of-the-art data workflow for processing CT datasets of electrode particle degradation. This method allows for large statistically relevant volumes to be analyzed during cell cycling whilst still tracking the changes in individual particles throughout time-lapse datasets. We believe that moving forward this workflow can be replicated for a broad range of electrochemical materials that observe volume and density change during use.