Prediction of rock-breaking forces of tunnel boring machine (TBM) disc cutter based on machine learning methods

Abstract

Rock-breaking forces are crucial indicators in evaluating the performance of tunnel boring machine (TBM) disc cutters, impacting cutter selection, cutterhead design, and penetration parameters. To develop an accurate and reliable prediction model for the rock-breaking forces of TBM disc cutters, a model database containing 414 typical samples was constructed based on the widely approved full-scale linear rock-breaking tests. The model takes the following inputs: uniaxial compressive strength (UCS), Brazilian tensile strength (BTS), cutter diameter (D), cutter ring tip width (T), penetration depth (P), and cutter spacing (S). The outputs are the normal and rolling rock-breaking forces. Four machine learning methods, i.e., back-propagation neural network (BP), support vector regression (SVR), K-nearest neighbors (KNN), and random forest (RF) were applied for the prediction model establishment and evaluation. Comparative analyses were performed against three well-known theoretical, semi-empirical, and empirical prediction formulas respectively. The results demonstrated that, despite the relatively small dataset, the predicted normal forces from BP, SVR, KNN and RF models achieved R-Square (R²) values of 0.82, 0.81, 0.77 and 0.89, respectively, significantly outperforming the other three prediction formulas. This confirmed the generalization and accuracy of machine learning algorithms. Among these models, the RF model showed the most stable predictive performance and was less sensitive to outliers. Further evaluations were performed using field penetration test data obtained from four TBM projects. Results showed that the machine learning models consistently achieved high prediction accuracy, whereas the three theoretical or empirical formulas were more affected by rock strength variations, and exhibited relatively poorer performance. The successful application and evaluation offered a valuable tool to assist TBM cutterhead/cutter design and operational parameters selection.