Convolutional Neural Network for Predicting the Performance of a Tunnel Boring Machine

The performance of tunnel boring machines (TBMs) is sensitive to rock mass properties, geological conditions, and machine parameters. Many researchers have applied traditional machine learning methods to predict the performance of TBMs over the past few decades. However, few research has yet been done on predicting the penetration rate (PR) with deep learning, which has proven to be effective in many fields in the past few years. In this study, a convolutional neural network (CNN) model is applied to predict the PR of TBM using field data from the fourth section of the Water Supply Project from Songhua River. The aim is to explore the relationship between the TBM performance and related parameters without rock mass properties. The mean absolute percentage error (MAPE) of the predicted PR by the CNN was 7.4%, while the conventional backpropagation neural network had a MAPE of 13.4%. These results show that the PR can be predicted accurately and reliably through the proposed CNN method.