CNN-Based 3D Characterization and Liberation Analysis of Lithium-Bearing Slag Particles Using Correlative CT and SEM Imaging

Efficient recovery of critical raw materials such as lithium from metallurgical slags requires optimized liberation of target phases during comminution. To determine effective mechanical process parameters for target phase recovery, an in-depth understanding of the characteristics of slag particles is crucial. For this purpose, modern tomography techniques, such as computed tomography (CT), can provide high-resolution 3D images of micrometer-sized slag particles. However, analysis of such CT images poses challenges, such as insufficient grayscale contrast between mineral phases and partial- volume effects. This paper presents a scalable workflow for accurate phase- and particle-wise 3D characterization of particle systems by correlating 3D CT images with 2D mineral maps. For this purpose, high-resolution scanning electron microscopy (SEM) slices are registered in 3D CT images and used as ground truth to train 3D convolutional neural networks (CNNs) for the segmentation of individual particles and mineral phases. This approach addresses the principal challenges of obtaining CT-based mineralogical characterizations, allowing for the particle-wise 3D characterization of complex slag systems with minimum manual labeling effort. The trained CNNs are then applied to CT images of particle systems with different particle sizes (from 63 μm to 100 μm and from 100 μm to 250 μm) of a lithium-bearing slag with LiAlO2 as the target phase. Although virtual cross-sections of the predicted 3D segmentations show excellent agreement with mineral liberation obtained from 2D validation SEM-EDS data, the derived 3D mineral liberation statistics differ significantly from 2D estimates. In particular, our results show that the 2D analyses significantly overestimate mineral liberation compared to the 3D characterization. By addressing this stereological bias, the correlative 3D characterization workflow provides essential insights required to tailor pyrometallurgical and mechanical processing parameters to improve the recovery of raw materials.