Abstract
The homogeneity of bentonite-graphite (B-G) mixtures is essential for the performance of engineered barrier systems in high-level radioactive waste (HLW) repositories. Uniform graphite distribution enhances thermal conductivity while maintaining the protective functions of bentonite. This study investigates the feasibility of using visible and near-infrared (VIS-NIR) spectroscopy combined with partial least squares regression (PLSR) to assess B-G mixture homogeneity. Small-quantity mixtures containing 0-5 wt% graphite were prepared, and hyperspectral data (400-1,000 nm) were collected. A PLSR model was developed to predict graphite content from the spectral data, achieving high accuracy (R² = 0.992, RMSE = 0.151 wt%) using five latent variables. The model was applied to large-scale mixtures blended using a double helix screw mixer. Samples collected at different time intervals were evaluated to determine the required mixing time for achieving homogeneity. After 120 min of mixing, statistical tests confirmed uniform graphite distribution. Buffer blocks were then fabricated using both homogeneous and inhomogeneous mixtures. Blocks from homogeneous mixtures showed no visible defects, whereas those from inhomogeneous mixtures exhibited edge spalling and lateral cracking. These results demonstrate that VIS-NIR spectroscopy with PLSR is a reliable, nondestructive method for evaluating mixture homogeneity and offers practical advantages for quality control in HLW buffer material production.