Detecting soil mixing, grain size distribution, and clogging potential of tunnel excavation face by classification-regression algorithms using EPBM operational data

Earth pressure balance machine (EPBM) operation is sensitive to the properties of the excavated soil due to the requirements of proper soil conditioning and maintenance of necessary chamber pressure. However, soil properties are invariably only available at a limited number of borehole explorations and soil samplings conducted during the subsoil investigation. Thus, it is crucial to identify properties of the tunnel excavation face, such as clay-sand mixed conditions, grain size distributions, and clogging potential along the whole alignment beside the few borehole locations to attain optimally efficient EPBM operation. Therefore, this paper presents the development of machine learning prediction models (i.e., classifiers and regressors) to estimate the properties of the tunnel excavation face using EPBM operational data collected during the tunneling operation as input features. Geotechnical data collected from boreholes and soil samples can be used to validate prediction models and calibrate them. To develop such models, the Northgate Link Extension (N125) tunneling project, constructed in Seattle, Washington, the USA, is used to capture and identify the true ground conditions. The results indicate successful prediction performances by the models, providing indication that EPBM parameters are crucial pointers of the tunnel excavation face properties to help attain optimally efficient EPBM operation.