Hyperspectral and image information of rock slag: An effective means to evaluate TBM tunneling efficiency

The characteristics of rock slags (including particle size distribution, geometric parameters, and elemental composition) are key indicators for evaluating the compatibility between tunnel boring machine (TBM) operational parameters and the surrounding rock conditions of tunnel face. Aiming at the limitations of traditional TBM tunneling efficiency evaluation methods and the lack of real-time feedback, this study constructs a tunnel excavation efficiency evaluation method based on image and hyperspectral technology. First, a You Only Look Once (YOLO) model optimized by ADown downsampling is used for rock slags image segmentation to obtain particle size distribution parameters. The established model achieves an accuracy of 99 % in identifying rock slags. Second, a Partial Least Squares Regression (PLSR) prediction model based on hyperspectral imaging technology is built to reverse the Fe content. The Competitive Adapative Reweighted Sampling- Partial Least Squares Regression (CARS-PLSR) model demonstrates the best inversion performance (R² > 0.96). Finally, by combining the obtained geometric features of the rock slags and Fe ion content, TBM tunneling efficiency and hob wear are predicted. The roughness index (CI) of rock slag is negatively correlated with the specific energy (R² = 0.869). When the thrust reaches 5.1 × 10³ kN, the roughness and particle size of the rock slag reach their peak values. Notably, higher rock slag iron content correlates with increased hob wear, suggesting that monitoring rock slag iron content can effectively predict hob wear. This method can provide a guarantee for the early warning of TBM safety construction in poor geology.