Tweet Share

• Send by email

  Send by email

  

  Applying machine learning for ground type detection using monitored earth pressure

  Geological uncertainty poses significant risk for soft ground tunnelling. The current ground mapping relies on a few boreholes to estimate the geology along the tunnel alignment via either expert judgment or geostatistical modelling, which is inevitably uncertain due to borehole sparsity. As earth pressure balance tunnel boring machine (EPBM) behaviour changes in various grounds and are measured during construction, it is possible to develop data-driven models to characterize the as-encountered ground condition while excavation. Using the Northgate Link Extension tunnelling project (N125) as a case study, it is shown that both supervised learning (SL) and semi-supervised learning (SSL) models can successfully characterize the as-encountered engineering soil unit fractions within the tunnelling envelope. Compared to prediction made by the geologist, SSL can take full advantage of the detail ring-by-ring data to yield realistic as-encountered ground condition. Compared to SL, SSL shows significantly better performance when insufficient data is available for training.

  Recipient :

  Name ^* :

  E-mail address ^* :

  ^* Required fields

  Send   or  Cancel

• Print