Reliability-based exploration: A quantitative method for evaluating and directing subsurface characterization.

摘要

A new computationally efficient method to quantitatively direct and evaluate the site exploration and characterization process was developed. This method, titled reliability-based exploration (RBE), compares the probability distribution of calculated performance to a predetermined performance goal to produce a reliability-index or probability of project success. The probability of project success is employed to evaluate the adequacy of site characterization data for a specific engineering project.; Three-dimensional engineering analyses and geologic characterization data are combined, without subjective judgment beyond acceptance of the initial subsurface model, to produce a variance in calculated performance, which is employed to quantitatively direct exploration. Variance in the calculated performance is computed through a first-order Taylor series expansion that incorporates spatially related sensitivity of the engineering analysis and spatially related uncertainty in geologic characterization data. Sensitivity of the analysis is calculated by direct derivative coding (DDC) of the three-dimensional finite element model (3-D FEM) and the meshing program. DDC allows sensitivities to be calculated with a single model run, thus eliminating the costly operation of repeatedly perturbing input variables to evaluate sensitivities. Because direct derivative modifications only effect the engineering analysis program and not the data, this programming technique for calculating sensitivities is not site or structure dependent and only needs to be undertaken once. Uncertainty in characterization data is computed through two different extrapolation approaches for comparison: kriging and conditional probability. During data extrapolation both site data and engineering judgement are incorporated into the RBE process.; The RBE method, while capable of including 3-D FEMs in general, is illustrated with several case studies where the performance measure is either total settlement or differential settlement and the uncertain characterization data are spatially varying geologic interface elevations. Settlement was selected as the performance measure because it is a fundamental geotechnical engineering issue; interface location was selected because its effect on engineering analyses is less investigated than that of soil properties.