Inverse analysis of a supported excavation through Chicago glacial clays

In inverse modeling, a model is calibrated by iteratively changing the estimates of its input parameters until the value of an objective function, which quantifies the errors between observed data and computed results, is minimized. The major advantage of an inverse analysis approach is the automatic and objective calculation of the parameter values that produce the best fit between observations and computed results. In this work model calibration is conducted using the optimization algorithm UCODE to update the design predictions of a supported excavation in soft clays using monitoring data collected during construction.; Results of triaxial compression tests on Chicago clay specimens are initially used to calibrate, by inverse analysis, four soil models: the Duncan-Chang, Modified Cam-Clay, Anisotropic Modified Cam-Clay and Hardening-Soil models. Parametric studies conducted for the MCC model on a number of regression variables show that the variables to which results are more "sensitive" are: (i) number and type of triaxial tests used as observations, (ii) number of input parameters estimated simultaneously and (iii) weighting of observations.; A numerical procedure that improves the state-of-the-practice of controlling ground movements associated with supported excavations is presented. The methodology, developed and tested using data from a 39 ft deep excavation through Chicago glacial clays, shows that: (i) inverse analysis based on field monitoring data can be effectively used to update the predicted performance of the supported excavation system, and (ii) successful recalibration of the model at an early construction stage positively affects subsequent "predictions" of the soil behavior. Results of a number of studies, conducted to evaluate the effect of modeling assumptions on the inverse analysis of the supported excavation, indicate that the three main characteristics of a "well-posed" inverse analysis problem are: effective numerical modeling, acceptable parameterization and appropriate choice of observations.