Optimazation of marine sediments characterization via statistical analysis
The task of geotechnical site characterization includes defining the layout of ground units and establishing their relevant engineering properties. This is an activity in which uncertainties of different nature (inherent, experimental, of interpretation…) are always present and in which the amount a...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Fecha de publicación: | 2022 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/369090 |
| Acceso en línea: | https://hdl.handle.net/2117/369090 https://dx.doi.org/10.5821/dissertation-2117-369090 |
| Access Level: | acceso abierto |
| Palabra clave: | Dynamic CPT(u) Semi-automated probabilistic tool Soil profiling Bayesian mixture analysis Hidden soil classes Soil unit weight Probabilistic landslide susceptibility map South west Iberian margin Cádiz, Golfo de |
| Sumario: | The task of geotechnical site characterization includes defining the layout of ground units and establishing their relevant engineering properties. This is an activity in which uncertainties of different nature (inherent, experimental, of interpretation…) are always present and in which the amount and characteristics of available data are highly variable. Probabilistic methodologies are applied to assess and manage uncertainties. A Bayesian perspective of probability, that roots probability on belief, is well suited for geotechnical characterization problems, as it has flexibility to handle different kind of uncertainties and highly variable datasets –in quality and quantity. This thesis addresses different topics of geotechnical site characterization from a probabilistic perspective, with emphasis on offshore investigation, on the Cone Penetration Test (CPTu) and on Bayesian methodologies.The first topic addresses soil layer delineation based on CPT(u) data. The starting point is the recognition that layer delineation is problem-oriented and has a strong subjective component. We propose a novel CPTu record analysis methodology which aims to: a) elicit the heuristics that intervene in layer delineation, facilitating communication and coherence in interpretation b) facilitate probabilistic characterization of the identified layers c) is simple and intuitive to use. The method is based on sequential distribution fitting in conventionally accepted classification spaces (Soil Behavior Type charts). The proposed technique is applied at different sites, illustrating how it can be related to borehole observations, how it compares with alternative methodologies and how it can be extended to create cross-site profiles. The second topic addresses strain-rate corrections of dynamic CPTu data. Dynamic CPTu impact on the seafloor and are very agile characterization tools. However, they require transformation to equivalent quasi-static results that can be conventionally interpreted. Up to now the necessary corrections are either too vague or require the acquisition of paired dynamic and quasi-static CPTu records (i.e., same location’s acquisition). A Bayesian methodology is applied to derive strain-rate coefficients in a more general setting, one in which some quasi-static CPTu records are available in the study area, but they need not be paired to any converted dynamic CPTu. Application to a case study offshore Nice shows that the results match those obtained using paired tests. Furthermore, strain rate correction coefficients and transformed quasi-static profiles are expressed in probabilistic terms.The third topic addressed is the optimization of soil unit weight prediction from CPTu readings. A Bayesian Mixture Analysis is applied to a global database to identify hidden soil classes within it. The goal is to improve the accuracy of regressions between geotechnical parameters obtained by exploiting the database. The method is applied to predict soil unit weight from CPTu data, a problem that has intrinsic practical interest but it is also representative of difficulties faced by a larger class of problems in geotechnical regression. Results highlight a decrease of systematic transformation uncertainty and an improve of accuracy of soil unit weight prediction from CPTu at new sites. In a final application we present a probabilistic earthquake-induced landslide susceptibility map of the South-West Iberian margin. A simplified geotechnical pixel-based slope stability model is considered to whole study area within which the key stability model parameters are treated as random variables. Site characterization at the regional scale combines a global database with available geotechnical data through a Bayesian scheme. Outputs (landslide susceptibility maps) are derived from a reliability-based design procedure (Montecarlo simulations) providing a robust landslide susceptibility prediction at the site according to Receiver Operating Curve (ROC). |
|---|