Integration of numerical modeling and field observations of deep excavations

In geotechnical engineering, precedent and observation of performance are an essential part of the design and construction process. For deep urban excavations designers rely on empirical data to estimate potential deformations and impact on surrounding structures. Numerical simulations are also employed to estimate induced ground deformations. Significant resources are dedicated to monitor construction activities and control induced ground deformations. While engineers are able to learn from observations, numerical simulations have been unable to fully benefit from information gained at a given site or prior excavation case histories in the same area.; A novel analysis method, self-learning in engineering simulations (SelfSim), is introduced to integrate precedence into numerical simulations. SelfSim extracts relevant constitutive soil information directly from field measurements of excavation response such as lateral wall deformations and surface settlement. The resulting soil model, used in a numerical analysis, provides correct ground deformations and can be used in the forward prediction of future excavations or later excavation stages. The soil model can continuously evolve using additional field information.; The feasibility of the framework is first demonstrated with synthetically generated field observations using the finite element method and two plasticity-based models (MCC and MIT-E3). Only surface settlements and lateral wall deflections from these analyses are used for SelfSim learning. The results show that the extracted soil model appears to have learned sufficient features of the soil behavior, in this case represented by MCC or MIT-E3, to reproduce the global response of the excavation.; The final chapter of this thesis includes the application of SelfSim to extract the soil constitutive behavior directly from field measurements of two excavation case histories. SelfSim is able to extract sufficient information on soil behavior to capture measured excavation response in multilayer soil profiles. The analyses demonstrate that the proposed SelfSim framework enhances our prediction and model learning capabilities from observed performance and represents a new opportunity to incorporate numerical simulations as an integral component in the application of the observational method in geotechnical engineering.