ABSTRACT
Recognizing the inherent challenges posed by the complex operational landscape of mechanized tunnelling by means of tunnel boring machines (TBM), and weighing in decades of experience in analysis of TBM data, extracting valuable insights from hundreds of kilometres of excavation data using conventional analytical and statistical methods, this contribution encompasses the frontiers of artificial intelligence (AI) in this domain.
This research presents an AI-powered anomaly detection tool for TBM data, utilizing an auto-encoding artificial neural network (ANN) to sift through the large amounts of information and identify hidden patterns. The ANN’s ability to learn and adapt makes it an invaluable tool for uncovering subtle anomalies that might be missed out using traditional methods, even in the face of complex operational factors such as tool wear, cutting disc clogging, and varying excavation conditions. By means of real TBM data from past projects, strengths and limitations of this AI approach are evaluated, acknowledging that even the most sophisticated algorithms can benefit from human expertise, especially when dealing with the intricacies of TBM operations. A hybrid system is proposed that seamlessly integrates AI with human judgment, enabling interactive data supplementation, expert-guided anomaly verification, and a deep understanding of the operational context.
The application example shows how the anomaly detection tool can be employed in different phases of a tunnel project and how even in early stages with a limited set of available data, engineers can benefit from automated anomaly indication.
