Piezocone Penetration Test (CPTu) is widely used in offshore projects to obtain soil parameters, such as the undrained shear strength. Due to depth limitations to perform this test, it is common to obtain data until around 40m when the conductor installation process would require at least double the depth. The present work uses extrapolation techniques based on analytic and heuristic approaches to estimate data beyond the depth domain of CPTu tests. Design of the conductor casing is highly dependent on soil properties, since it serves as a foundation element of the well. Estimation of the soil parameters is based on deepwater CPTu data from Brazilian offshore basins. Three analytical approaches are used in this study (linear and non-linear regressions: second and third degrees). Moreover, Artificial Neural Networks (ANN) (dense, convolutional and recurrent networks) are also employed to predict the soil behavior. Methodologies were applied, validated and compared to evaluate their capability to accurately estimate the undrained shear strength. Python subroutines were developed and applied to sets of homogeneous and heterogeneous data from CPTu tests. The undrained shear strength was estimated beyond the test domain until the depth of interest to the conductor casing design around 80m. For this purpose, both groups of techniques were validated analyzing the efficiency of the fitting process, the associated error and coefficient of determination of each methodology. From that point on, we compared data from analytical methods and the neural networks application, verifying which technique fits better on the datasets. These methods of estimation of soil properties work as an instrument to support the decision-making process in top-hole drilling operations. The datasets analyzed present different levels of soil heterogeneity and performing the extrapolation analysis brings additive information to understand the soil behavior beyond depths reached by CPTu tests. This contributes to the safety and reliability of conductor casing design and installation. To the authors' knowledge, this analysis is rarely performed in the literature.

No abstract

Summary Evaluation of characteristic values of geotechnical parameters (used for oil well design) is associated with uncertainties inherent to the geological processes that change soil strata. Statistical analysis of soil data allows one to deal rationally with these uncertainties. This work addresses some normative recommendations and literature models for statistical characterization of undrained shear strength and submerged unit weight in offshore soils, providing more information to conductor casing design. Three methodologies were selected for the analysis. The NORSOK G-001 (2004) standard recommends using mean values computed conservatively. Lacasse et al. (2007b) propose that the characteristic value should be the mean value minus one-half of a standard deviation of the parameter under analysis. DNV-RP-C207 (2012) suggests different methodologies for dependent and independent soil variables, though both methods of calculating the characteristic value involve linear regressions. Using data from geotechnical investigations that characterize eight oil wells located in two Brazilian offshore basins, the selected methodologies were applied to obtain the characteristic values and compared to each other. The analysis is carried out with data from 17 piezocone penetration tests (CPTu) associated with the eight wells mentioned above. It is noted that the NORSOK recommendation leads to the highest characteristic values, which are assumed tending to the mean value of the data set over the well depth. The values obtained using Lacasse et al. (2007a, 2007b) methodology are more conservative than NORSOK methodology and stand as its lower bound. The models suggested by DNV perform differently when applied to the geotechnical parameters. The dependent variables methodology fits both undrained strength and unit weight accurately. Analysis shows that undrained strength is better described using the methodology for standard deviation proportional to the depth, while for the unit weight, accurate results are obtained by using constant standard deviation. The lower bound procedure proposed by DNV provides, in general, higher results in the first meters and more conservative values along the depth when compared with the other methodologies. Regarding all the formulations addressed, differences between them increase for wells whose CPTu data present higher dispersion. This larger dispersion suggests applications of different statistical-based approaches in order to reliably characterize offshore soil data. The data sets analyzed comprise different levels of scattering and soil heterogeneity, and comparing the statistical recommendations brings additive information for the designer to set the characteristic values of soil properties, aiming for the decision-making process on top hole drilling applications (e.g., conductor casing design). To the authors’ knowledge, few papers perform this comparative analysis.

Evaluation of characteristic values of geotechnical parameters is associated with uncertainties inherent to the geological processes that change soil strata. Statistical analysis of soil data allows one to deal rationally with these uncertainties. The present work addresses some normative recommendations and literature models for statistical characterization of undrained shear strength and submerged unit weight in offshore soils, providing more information to conductor casing design. Three methodologies were selected for the analysis. The NORSOK G-001 (2004) standard recommends the use of mean values computed conservatively. Lacasse et al. (2007) propose that characteristic value should be the mean value minus half a standard deviation of the parameter under analysis. DNV-RP-C207 (2012) suggests different methodologies for dependent and independent soil variables, though both methods of calculating the characteristic value involve linear regressions. Using data from geotechnical investigations that characterize eight oil wells located in two Brazilian offshore basins, the selected methodologies were applied to obtain the characteristic values and compared to each other. The analysis is carried out with data from seventeen CPTu tests associated to the eight wells abovementioned. It is noted that the NORSOK recommendation leads to the highest characteristic values, which are assumed tending to the mean value of the dataset over the well depth. The values obtained using Lacasse et al. (2007) methodology are more conservative and stand as a lower bound of the NORSOK methodology. The models suggested by DNV perform differently when applied to the geotechnical parameters. The dependent variables methodology fits both undrained strength and unit weight accurately. Analysis shows that undrained strength is better described using the methodology for standard deviation proportional to the depth while, for the unit weight, accurate results are obtained by using constant standard deviation. The lower bound procedure proposed by DNV provides, in general, non-conservative results in the first meters and more conservative values along the depth when compared with the other methodologies. Regarding all the formulations addressed, differences between them increase for wells whose CPTu data present higher dispersion. This larger dispersion suggests applications of different statistical-based approaches in order to reliably characterize offshore soil data. The datasets analyzed comprise different levels of scattering and soil heterogeneity, and comparing the statistical recommendations brings additive information for the designer to set the characteristic values of soil properties, aiming for the decision-making process on top hole drilling applications, e. g. conductor casing design. To the authors’ knowledge, few papers perform this comparative analysis.