Name: GABRIELE CARVALHO BAHIENSE CARIAS
Publication date: 25/03/2024
Examining board:
Name |
Role |
|---|---|
| BERNADETE RAGONI DANZIGER | Examinador Externo |
| HERALDO LUIZ GIACHETI | Examinador Externo |
| KATIA VANESSA BICALHO | Presidente |
| MARISTELA GOMES DA SILVA | Examinador Externo |
| WILIAN HIROSHI HISATUGU | Coorientador |
Summary: The results of field tests for determining soil resistance to penetration are essential for defining geotechnical behavior and estimating parameters in civil engineering projects. Due to the variety of tests and the impossibility of complete investigations, it is necessary to evaluate correlations between different tests, considering their specific characteristics. This research analyzed the influences on the correlations defined between the results of SPT and CPT (and their variations, such as piezocone penetration test, CPTu) soil resistance determination tests, and examined the use of artificial neural networks to correlate field test results for determining soil resistance to penetration. A statistical pre-treatment of the data was performed, evaluating the frequency distribution of the different analyzed parameters and their ranges of variation. The study also discussed the benefits obtained by considering the results of DMT tests associated with CPTu and SPT tests for identifying soil mechanical behavior, highlighting
interpretation divergences when considering the three in-situ tests. Additionally, the influence of parameters such as depth, lateral friction, excess pore pressure, among others, on the correlations defined between the SPT and CPTu soil resistance determination test results was evaluated. An artificial neural network (ANN) model of the multilayer perceptron (MLP) type was developed to correlate in-situ penetration test results. The data used in this research consisted of 38355 sets of SPT-CPT-DMT results for sandy, silty, and clayey soils. The input variables of the ANN model include: SPT N60 penetration resistance index, cone tip resistance
(qt), lateral friction (fs), pore pressure (u2), and material index (Ic) from the CPTu, as well as the material index (ID) from the DMT. The research demonstrates that factors such as depth and lateral friction influence the SPT-CPTu relationships, and neural networks trained only with qt and N60 showed low performance. However, the inclusion of depth, lateral friction, and the DMT ID improved the statistical performance of the model, which, nevertheless, does not fit a generic equation.
