The shield tunneling machine is a critical piece of equipment in tunnel construction, renowned for its high efficiency, safety, and stability. Nonetheless, the occurrence of boulders, particularly in coastal regions, presents substantial challenges to excavation activities. The timely and accurate detection of boulders is essential for the establishment of effective warning systems. Given the limited sample size of boulders and the complexities associated with big data mining, the real-time detection of boulders in coastal tunnels is a formidable task. This study introduces an innovative hybrid machine learning (HML) model for the detection of boulders. The research is grounded in data from the Shenzhen Metro project and encompasses extensive data preprocessing, leveraging equipment data and an extensive boulder sample library. This method integrates feature extraction, unsupervised learning, and ensemble learning modeling techniques and systematically assesses their impact on predictive accuracy. The findings of the study indicate that sophisticated feature extraction methods are instrumental in enhancing the precision of giant rock prediction. Unsupervised clustering has demonstrated its efficacy in classifying data samples based on their structural similarity, thereby refining prediction accuracy through the generation of refined labels. The stacking method outperforms traditional techniques by employing a two-tiered learning approach, exhibiting superior predictive capabilities. Furthermore, in engineering applications, this study offers a comprehensive solution for the automated detection of boulders in coastal shield tunnel construction. By incorporating the HML model, this study has enhanced the capabilities of shield tunneling equipment, paving the way for safer and more efficient tunnel excavation processes.
Xiao et al. (Thu,) studied this question.