SOIL BEHAVIOR MODELLING USING CPTU DATA AND NEURAL NETWORKS FOR OFFSHORE OIL WELL DESIGN

Resumo

Piezocone Penetration Test (CPTu) is widely used in Petroleum Engineering for soil profiling
and estimation of soil parameters (undrained shear strength, submerged weight), which are essential
geotechnical parameters used in the design of oil wells. The in situ test results are used in several
techniques for soil characterization, aiming the design of the conductor casing, which is the tubular that
provide structural support to the well, serving as its foundation element. Despite its efficiency, this test
bears some logistic limitations regarding depth of analysis since CPTu data is usually obtained within a
range of 40% to 50% of the conductor casing drilling depth. Data of undrained shear strength of the soil
beyond the depths covered by CPTu tests are beneficial to the safety of the drilling operation. Due to
the natural variability of the materials, the evaluation technique of this soil property must consider soil
heterogeneity as a premise. The present work uses artificial neural networks (ANNs) to predict undrained
shear strength behavior in offshore soil for depths not reached by CPTu equipment. The MLP (Multilayer
Perceptron) networks are trained with test data of different types of soil from two Brazilian offshore
basins, using classification techniques to define soil strata and segment the soil response estimation.
The error is evaluated using cross-validation procedures for different proportions of training/prediction
data. The expected results include the definition of the network architecture and prediction accuracy in

different types of soil strata. This kind of study on the soil characterization aims to support the decision-
making process on well casing design, allowing a robust structural well integrity analysis.