Técnicas de inferência e previsão de dados como suporte à análise de integridade de revestimentos

Abstract

The design of an oil well is a complex and multidisciplinary activity, which has as one of its main premises the adequate prediction of the well’s integrity throughout its life cycle. Despite all the effort taken in designing its structure, eventually, the well may be exposed to unpredicted loading conditions. Regarding the well monitoring, the acquisition of data referring to variables such as pressure and temperature allows identifying whether the well is operating within the parameters predicted in the design. By using time series forecasting techniques, the information gathered from the sensors can be used to diagnose an anomaly that has already occurred and prevent its occurrence, creating a real-time supervisory system capable of anticipating future loading states. Furthermore, in the context of well design, with the evolution of the rules for casing design, it is suggested to use probabilistic methods, highlighting the importance of better knowledge about the design variables. The statistical inference about manufacturing data is motivated by the demand for a better understanding of these parameters’ uncertainties in terms of the dimensions of the tubular and the properties of the metallic alloy that constitutes it. This dissertation approaches a set of statistical inference and data prediction techniques, aiming to support designing practices and monitor the structural integrity of wells.