Modern earth pressure balance tunnel boring machines (EPBMs) are extensively instrumented to gather data about human operations, machine reactions and construction logistics. These readily available data are the most direct, high resolution and unbiased description of the tunneling process, containing rich information on the physics of tunneling and the human aspect of operation. However, these data are underutilized in current tunneling practice, motivating the development of new approaches to make use of EPBM data. This thesis addresses this challenge by proposing three applications to achieve improved understanding and control of EPBM tunneling based on EPBM data. First, using the EPBM data recorded at an active tunnel site, a simplified chamber pressure model is developed, where two physical processes are considered and modeled: compressible material flow and chamber fluid seepage. The former treats the muck as a three-phase mixture whose behavior is void (and hence pressure) dependent and changes with the applied EPBM operations (i.e., screw conveyor rotation, soil conditioning and EPBM advancement), while the latter models the pressure dissipation during standstill using Darcian flow. A case study of the N125 tunneling project suggests the model is capable of tracing the chamber pressure variation in tunneling. However, with the increase of simulation duration, its accuracy degrades due to the accumulation of error. A model sensitivity study also suggests that the screw conveyor efficiency is a critical model parameter, whose value is shown to correlate with the formation soil water content and soil conditioning agent volume used. Second, using EPBM data and limited borehole logs, a framework is developed to characterize the as-encountered ground condition to allow for improved awareness of ground conditions while tunneling. Each ring is represented by features developed from the EPBM data to predict the ground condition, which is simplified to engineering soil unit (ESU) fractions inside the tunneling envelope, encoded probabilistically from borehole logs. Applying both supervised learning (SL, with multinomial logistic regression) and semi-supervised learning (SSL, with label propagation), the results suggest the major ESUs can be successfully identified, with the average Kullback-Leibler divergence (DKL) of the SL and SSL methods being 0.64 and 0.51, respectively. The average precision/recall are 0.62/0.60 for the SL method and 0.66/0.63 for the SSL method. The average F1-scores are 0.61 for the SL method and 0.64 for the SSL method. Compared to the inference made by the geologists using borehole logs alone, the ESUs detected by data-driven approaches are more heterogeneous and can locate the boundaries between ESUs with finer resolution. Besides, the comparison between SL and SSL results conveys that while their performances are similar given sufficient borehole logs for training, SSL performs significantly better than SL when limited borehole logs are available for training. The study also shows that visualizing the similarity graph, the byproduct of the SSL method, offers an intuitive way for model interpretation. Third, this thesis also presents an effort of applying data-driven optimal control to enhance the EPBM tunneling performance. Using support vector regression, two coupled physical processes are modeled: EPBM advancement and cutterhead rotation torque. These models are incorporated into an optimal control framework and are optimized to determine the operations yielding the maximal instantaneous tunneling advance rate (AR). The results suggest an average AR increase by 8.6% and an average cutterhead torque reduction by 4.6% can be achieved. It is also found that the optimal thrust force and the resulting AR improvement are influenced by the chamber pressure, with a lower pressure magnitude requiring a smaller thrust force and thus making a higher AR achievable. The optimal cutterhead rotation speed, however, is found to be independent of the chamber pressure, and faster cutterhead rotation always contributes to more rapid EPBM advancement.
Copyright of the original work is retained by the author.
The export option will allow you to export the current search results of the entered query to a file. Different
formats are available for download. To export the items, click on the button corresponding with the preferred download format.
By default, clicking on the export buttons will result in a download of the allowed maximum amount of items.
To select a subset of the search results, click "Selective Export" button and make a selection of the items you want to export.
The amount of items that can be exported at once is similarly restricted as the full export.
After making a selection, click one of the export format buttons. The amount of items that will be exported is indicated in the bubble next to export format.