3D Reconstruction and Analysis of Lithium-ion Batteries from X-ray Images

The three-dimensional reconstruction of lithium cobalt oxide (LCO) batteries in the X-ray domain is crucial for analyzing their internal structure and degradation processes. Traditional reconstruction techniques require acquiring and aligning thousands of X-ray projections, making the process computationally expensive and impractical for certain applications. In this work, we explore state-of-the-art rendering techniques to significantly reduce the number of required projections while maintaining high reconstruction quality. Our primary focus is on 3D Gaussian Splatting (3DGS), which we demonstrate to be a faster and more efficient alternative compared to Neural Radiance Fields (NeRF). We begin by introducing the problem, detailing the X-ray acquisition system, and discussing the role of X-ray absorption fine structure (XAFS) technology in LCO battery analysis. Through systematic experimentation, we compare 3DGS and NeRF, showing that 3DGS achieves superior reconstruction quality with significantly lower computational cost and training time. Additionally, we investigate strategies for optimizing reconstruction quality by carefully selecting, averaging, and filtering projections to minimize acquisition and processing overhead. Our findings confirm that modern rendering techniques, particularly 3DGS, enable high-fidelity 3D reconstructions with an order-of-magnitude reduction in required projections, making them a promising solution for efficient battery analysis.