Deep Learning-Based Image Classification of 18650 Lithium-Ion Battery Structural Health Using X-ray Micro-Computed Tomography

Lithium-ion batteries (LIBs) are ubiquitous in modern technology, powering consumer electronics, electric vehicles, and energy-storage systems. As these systems age, internal structural degradation can lead to reduced performance, diminished lifetime, and increased safety risks, including thermal instability. Because many forms of degradation occur internally and are not detectable through external measurements, accurate assessment of structural health can be observed by non-destructive imaging and robust analysis techniques. In this study, a transfer learning-based deep learning framework for classifying the structural health conditions of 18650-format LIB cells using X-ray micro-computed tomography (µCT) imaging is proposed. This approach includes preprocessing that extracts radial CT slices and core-region cropping to capture localized 3D structure. The dataset is balanced and augmented with transformations and rotations, and a pretrained InceptionResNet-V2 model is fine-tuned to distinguish between various cell conditions. Modified classification layers with dropout and class weighting improve robustness. Initial results demonstrate that the model can identify internal structural differences with promising accuracy, supporting the development of automated µCT-based battery health assessment and safety diagnostics.