为实现更精确、更实时的智能围岩分级以匹配双护盾TBM高效掘进，基于先进的HRNet高分辨率神经网络，通过分析地质与掘进参数2类指标的强关联性，构建基于该2类指标融合的高分辨率TBM隧道双模态围岩精准智能分级模型，开展实时的围岩分级工作。结果表明： 1）推进速度与刀盘转矩2个参数与围岩坚硬度有较强的关联性，通过这2个掘进参数二维组合的线性趋势以及拐点区间可以区分不同的岩石坚硬程度； 2）刀盘滚动角和撑靴俯仰角2个参数与岩体完整性有较强的关联性，通过这2个掘进参数的摆动幅度和角度可区分不同的岩体完整性； 3）通过将2个模态单一的HRNet高分辨率神经网络进行拼接和全连接，构建的双模态围岩分级模型能够实现离散型数值的地质模态数据与连续型数值的掘进模态数据融合，该模型能够从围岩稳定性与岩机交互等多方面更加精确地反映围岩分级，应用准确率可达95.0%； 4）相比仅用掘进指标的单模态模型，双模态模型在测试集上表现效果更好，其Macro F1 Score为0.985。
Efficient excavation and closed construction
environment of double-shield TBM relies on more accurate
and real-time intelligent surrounding rock
classification methods. To achieve this, the strong correlation between the two
indices of geology and excavation parameters is analyzed based on advanced
HRNet high-resolution neural networks. Furthermore, an
accurate dual-mode intelligent classification model of
surrounding rock with high-resolution is built based on
the fusion of these two indices. As a result, real-time
surrounding rock classification can be conducted. Main findings are as follows:
(1) The advance speed and cutterhead torque are strongly correlated with rock
hardness. The linear characteristics of these two parameters distinguish
different rock hardness. (2) The cutter rotation angle and gripper pitch angle
are strongly correlated with rock integrity, and different rock integrity can
be distinguished by these two parameters. (3) A dual-mode
classification of surrounding rocks established by bonding and fully connecting
two single-mode HRNet high-resolution
neural networks fuses discrete numerical geological modal data with continuous
numerical tunneling modal data, thereby more accurately reflecting the
surrounding rock classification from multiple aspects such as surrounding rock
stability and rock-machine interaction. The application
accuracy reaches 95%. (4) The dual-mode model
outperforms single-mode model on testing set, with an
Macro F1 Score of 0.985.
