Reliability-Based Numerical Analysis of Spread Foundations Incorporating Spatial Variability on a Complex Substratified Layered Medium

Abstract

The modern partial factor design is paired with a number of diverse scenarios to be considered on geotechnical reliability-based design. A study regarding a reliability-based numerical analysis is presented and results are provided on the analysis of a strip spread foundation designed by the Eurocode 7 methodology. The first design example on a relatively homogeneous ground is supplemented by a second design example on a layered medium wherein are equivalent the mean shear strength parameters of the foundation soil weighted in the interested region for bearing capacity safety assessment and the parametric average along the geotechnical profile. On the third design example the shear strength parameters of the foundation soil are implemented to account for the spatial variability on a complex substratified layered medium wherein a number of heterogeneous clusters are assigned in a nonsymmetric calculation. The neural networks approach is then tested as a metamodelling technique considered the possibility to capture the nonlinear interactions in a system. The technology is paired with other polynomial techniques based on a regression approach and consists on a mechanism able to compute a mapping on the multivariate space given a set of data. The applicability of neural networks is as well highlighted in the context of fitting and pattern recognition, namely on the classification of failure modes. A limit state probabilistic analysis for safety assessment is then presented in the format of a sensitivity analysis to identify probability bounds for scenarios of interest, considered the intervals on the probabilistic evaluation.