Physics-informed and data-driven machine learning of rock mass classification using prior geological knowledge and TBM operational data

Predicting rock mass classifications ahead of tunnel faces is essential for tunnel construction by full-face hard rock tunnel boring machine (TBM), because such an accurate prediction of rock mass classifications provides timely and valuable information for decision-making by TBM drivers and practitioners to minimize tunneling accidents and improve construction efficiency. Previous studies have demonstrated the values of TBM operational data in predicting rock mass classification. Several purely data-driven machine learning (ML) methods have been developed to capture the complex relationship between TBM operational parameters and rock mass classes ahead of tunnel faces. However, these existing ML methods are purely driven by TBM operational data, without using other available information, such as prior geological knowledge that has been collected from site investigation during planning and design stages of the tunnel projects before TBM construction. To leverage on the available prior geological knowledge, a physics-informed and data-driven ML model is proposed in this study to predict the rock mass classifications ahead of tunnel faces. This physics-informed and data-driven ML model integrates prior geological knowledge collected from site investigation and TBM operational data collected during the tunneling process. The proposed method is applied to predict rock mass classification at the Songhua River water conveyance project in China and performs well. © 2024 Elsevier Ltd.