The rapid expansion of electric vehicles has accelerated the need for advanced battery recycling technologies that ensure high resource recovery efficiency and economic sustainability. Among various approaches, direct recycling has gained significant ...
The rapid expansion of electric vehicles has accelerated the need for advanced battery recycling technologies that ensure high resource recovery efficiency and economic sustainability. Among various approaches, direct recycling has gained significant attention because it allows recovery of cathode active materials while preserving their structure, enabling reuse without resynthesis. However, accurately assessing the degradation state of electrodes during the re-lithiation step remains challenging, and as a result, re-lithiation is typically carried out with an excess amount of lithium. This leads to the formation of residual lithium species such as Li₂CO₃ and LiOH, which generate gas and consequently degrade cell performance, thereby requiring further post-processing, such as washing or heat treatment. Therefore, a nondestructive and reliable method for diagnosing electrode State of Health (SOH) is essential.
In this study, X-ray Microscopy (XRM) was used to obtain high-resolution, nondestructive images of NCM cathodes cycled under three different conditions, with samples harvested at specific SOH levels for each condition. Because structural differences across SOH levels were not clearly distinguishable through qualitative observation, a quantitative image-analysis pipeline was developed. This pipeline includes K-means clustering for component segmentation, local watershed–based particle identification, feature extraction at both image and particle levels, and feature importance evaluation using mutual information and random forest–based metrics. The selected features were used as inputs to the SOH classification model, which was independently trained and evaluated for each of the three cycling conditions (25 °C 4.165 V, 45 °C 4.165 V, 25 °C 4.2 V).
The proposed model achieved SOH classification accuracies of 95%, 83%, and 80% for each condition, with an average accuracy of 86.1%, demonstrating a substantial improvement over qualitative or conventional image-based diagnostic approaches. These results indicate that XRM-derived structural information can effectively reflect electrode degradation through appropriate quantification and feature selection.
The diagnostic approach presented in this work can support more informed lithium adjustment strategies in direct battery recycling by providing reliable SOH information, which may help reduce the formation of residual lithium species and ease the burden of downstream processes such as washing or heat treatment. This nondestructive SOH diagnostic methodology contributes to the advancement of battery recycling technologies and enhances their practical applicability.