Prediction of geological conditions for a tunnel boring machine using big operational data

Abstract

This paper presents a comprehensive procedure to predict geological conditions (i.e., rock mass types) for a tunneling boring machine (TBM) based on big operational data including four channels: cutterhead speed, cutterhead torque, thrust, and advance rate. To handle the big operational data, a Balanced Iterative Reducing and Clustering using Hierarchies (BIRCH) algorithm is adopted to effectively compress 12,038,636 TBM operational data to only 5014 leaf node entries. A K-means++ algorithm is used to find potential rock mass types in the TBM operational data. By comparing three kinds of classifiers, a support vector classifier (SVC) with an average precision of 98.6% is selected as the best geological conditions prediction model. Test results on historical TBM operational segments show that most adjacent operational segments have the same rock mass type. The change in rock mass type is a dynamic process, which first fluctuates between two rock mass types and gradually stabilizes at the latter type. In addition, the cutterhead torque and thrust are found to better reflect the change of rock mass types compared with the advance rate and cutterhead speed. Test results on a water conveyance tunnel show that using only 20% of TBM training data the developed prediction model can generate 84.4% precision and 88.8% recall performance for the remaining 80% testing data. Hence, the proposed procedure could be applied to big TBM operational data to accurately detect, characterize, and predict rock mass types, which is of critical importance to safe and efficient tunneling.