Análise estatística em dados de perfilagem como suporte à avaliação da integridade do revestimento de poços de petróleo

Abstract

A well casing consists of a tubular structure in high-strength steel responsible for different functions, during well construction phase and also in its lifespan. During drilling operation of a well, casing wear, caused by contact between the drilling column and the inner wall of the casing is a problem that directly impacts tubulars structural integrity. This phenomenon is difficult to predict and causes a reduction in tubulars resistance, which together with the extreme conditions of the well, increases the risk of structure failure. In an attempt to assess structural integrity of executed wells, the oil and gas industry uses sensors in a technique called ultrasonic logging, applied after drilling operations. Ultrasonic logging provides casing thickness and internal radius data along casing column, providing means to identify structural damage by casing wear. In this work, a statistical analysis is applied, in addition to numerical fitting methods, to logging data, in order to improve casing wear identification in casing cross sections. Least square method is applied in order to fit a theoretical distorted and rotated elliptical shape to ultrasonic logging data. The fitting procedure enables quantify casing wear and also captures geometrical imperfections like ovality and eccentricity presents in data. The statistical analysis is applied to residuals from fit and consists of strategies developed to identify the statistical parameters of the noise present in logging devices, which makes it possible to better assess the geometry variations in the investigated sections and provide greater reliability in the quantification of casing wear. Goodness of fit tests are applied, together with the statistical noise parameters, and also compared, for the study and interpretation of logging data. The proposed methodologies are verified through simulation using virtual sections generated with wear, noise and known geometry. The verified data treatment methodology, presented in this work, are also applied to a case study with real logging data.