MMIJ Annual Meeting

Accurately assessing the mechanical properties of the surrounding rock mass is crucial for ensuring safety and stability during the construction of underground tunnels. Although the use of measurement while drilling (MWD) data in tunneling projects is on the rise, further exploration of machine learning (ML) techniques for the real-time identification of geological conditions in the surrounding rock mass is still required. This paper introduces a novel ML framework for identifying the strength of drilled rock using MWD data. To assess the framework’s performance, an indoor MWD test was conducted. The study applied various ML algorithms, including the XGBoost model, to identify rock strength based on four key MWD features: torque, drilling thrust, drilling speed, and rotational speed. A selection of the most popular parameter optimization algorithms was explored to enhance the prediction accuracy of the ML models. The rock strength identification framework presented in this paper follows a rigorous scientific ML modeling process. The test results demonstrate the significant potential of using optimization-enhanced ML models for accurately identifying rock strength.