Dynamic Cutterhead Torque Prediction of Tunnel Boring Machines Based on a Hybrid Transfer Learning Framework

Abstract

AbstractDynamic cutterhead torque prediction plays an essential role in the design and safe operation of a tunnel boring machine (TBM) system, since it can be used to assist equipment health diagnosis and energy consumption estimation. Complicated and variable geological conditions, leading that operational and status parameters of TBMs usually exhibit some spatio-temporally varying characteristic, which poses a serious challenge to data-based methods for dynamic cutterhead torque prediction. In this paper, a novel hybrid data mining (DM) framework based on clustering, multitask learning (MTL), transfer learning, and least-squares support vector regression machines (LS-SVR), abbreviated as TRLS-SVR, is proposed for cutterhead torque prediction of TBM. In the TRLS-SVR, the MTL paradigm based on the minimization of regularization function similar to LS-SVR, was applied to exploit reliable and representative knowledge from long-ago historical data. With the application of transfer learning, knowledge learned from historical dataset is retained and leveraged to train a new model for current prediction. A collection of heterogeneous in-situ TBM data from a tunnel project located in China is utilized to evaluate the performance of the proposed framework. Experiment results validate that the prediction performance of the proposed framework outperforms that of existing data-driven prediction methods.KeywordsTunnel boring machines (TBMs)Cutterhead torque predictionOperation parameterTransfer learning