Online Geological Anomaly Detection Using Machine Learning in Mechanized Tunneling

In the case of a sudden change in the geology in front of the Tunnel Boring Machine (TBM) during mechanized tunneling processes, non-appropriate investigation and process adaptation may result in non-desirable situations that can induce construction and machine defects. Therefore, subsurface anomalies detection is necessary to trigger alarm to update the process. This paper presents an approach for geological anomaly detection using data produced by the TBM. The data observations are continuously produced at a motion of 10 to 15 s from hundreds of sensors around the TBM. Unsupervised machine learning techniques are applied to analyze the online streaming data. As a result, a model, which is able to learn the system characteristics from normal operational condition and to flag any unanticipated or unexpected behavior, is established. The proposed approach has been tested on the data of the Wehrhahn-Linie metro project in Dusseldorf in Germany. The model can accurately detect the presence of concrete walls in the ground domain with a distance up to around one meter before the TBM approaches the walls. The developed method can thus be used as a monitoring system for ground risks detection to ensure safe and sustainable constructions in mechanized tunneling.