Unsupervised spectral clustering for shield tunneling machine monitoring data with complex network theory

Extraction of underlying knowledge from monitoring data is beneficial to on-site management during shield tunneling construction. However, it remains a challenge to unsupervisedly learn from these monitoring data due to the nature of high-dimensional, miscellaneous, and highly nonlinear. This study proposes a new systematic approach to classify shield machine monitoring data by integrating spectral clustering (SC) and complex network (CN) theory. In this approach, CN theory is introduced to obtain the topological relations and network structure of machine monitoring data directly. Based on the network topology, SC is employed to classify these data with unbiased similarity measurement. A river-crossing shield tunnel is used in this study to validate the effectiveness and feasibility of the proposed approach. It's demonstrated that the proposed approach outperforms the other SC methods for unsupervised classification of the machine monitoring data. The classification of machine monitoring data with the proposed approach has a potential value in machine performance and geological risk assessment during shield tunneling construction.