A real-time rock mass class identification model of the tunnel face based on TBM tunneling and the corresponding muck characteristic parameters

Abstract

TBM tunneling is the result of interaction between the rock mass and the machine. Rapid identification of the rock mass condition at the tunnel face is crucial for the safety and efficiency of TBM tunneling. This study was based on the spiral ramp project of the Beishan Underground Research Laboratory. A TBM muck analysis system was installed on the TBM conveyor belt to obtain the muck characteristic and TBM tunneling parameters. Combining the muck characteristic parameters, TBM tunneling parameters and the corresponding rock mass classes at the tunnel face, a multi-source database was established. Subsequently, machine learning models for rock mass class identification were developed based on TBM tunneling parameters, muck characteristic parameters, and their fusion, respectively. The LightGBM model based on these fusion parameters including tunneling and muck characteristic parameters, significantly outperforms other models, achieving an Accuracy of 0.934, an F1-score of 0.932, and a Kappa coefficient of 0.904. The model was validated in the subsequent TBM tunneling in the same project. It demonstrated the reliability of the model in practical applications.