Bayesian mixture analysis of a global database to improve unit weight prediction from CPTu
Empirical correlations between different geotechnical parameters are frequently sought after by exploiting multivariate databases. An example are estimations of total unit weight from cone penetration tests (CPTu), which are very useful in earlier design stages. If the underlying soil database inclu...
| Autores: | , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2023 |
| 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:2117/403965 |
| Acceso en línea: | https://hdl.handle.net/2117/403965 https://dx.doi.org/10.1016/j.enggeo.2023.107353 |
| Access Level: | acceso abierto |
| Palabra clave: | Soil penetration test BMA Total soil unit weight Cone penetration test Systematic uncertainty Global correlations Mecànica dels sòls -- Prospeccions i sondatges Àrees temàtiques de la UPC::Enginyeria civil::Geotècnia::Mecànica de sòls |
| Sumario: | Empirical correlations between different geotechnical parameters are frequently sought after by exploiting multivariate databases. An example are estimations of total unit weight from cone penetration tests (CPTu), which are very useful in earlier design stages. If the underlying soil database includes data from many different soils a single correlation may lack precision. Precision is gained when the underlying database is narrowed down to some specific soil type, but the applicability of a soil-specific correlation is also limited. A way out of this dilemma is to apply clustering techniques to a general database before developing separate correlations for different clusters. Projecting the clustered data back to a convenient classification space (e.g., one spanned by normalized CPTu metrics) new data can be easily assigned to different clusters and the appropriate correlation used. This idea is illustrated here using Bayesian Mixture Analysis (BMA) to identify hidden soil classes within a general geotechnical database that supports correlations between soil total unit weight and CPTu readings. It is shown that BMA supported clustering improves the accuracy of previous regressions, and, more importantly, facilitates the formulation of novel and more accurate regressions. A simple discriminant criterion is developed to facilitate application of cluster-based regressions to new sites. The good performance of the method is illustrated with application to a deltaic site. |
|---|