Exploratory Analysis of the 3W Dataset for Detecting Operational Failures in Oil Wells Using Machine Learning Techniques

In the oil and gas industry, offshore operations are subject to environmental regulatory requirements, especially in exploration and production. These requirements aim to minimize environmental impacts and ensure operational safety, particularly in well management. Flow control and the safety of the extraction system are monitored due to the risks associated with any type of failure. Given the complexity and inherent risks of offshore-well extraction operations, the oil industry seeks methods and technologies that allow the prediction and anticipation of operational failures. The development of monitoring and failure prediction techniques aims to reduce the number of incidents and accidents, ensuring production continuity and minimizing losses associated with unplanned shutdowns. In the work of JARDINE et al. (2006), the concept of condition-based monitoring (CBM) is discussed, a maintenance program that recommends maintenance decisions based on collected information. This is a system that supports predictive maintenance decisions. In 2019, PETROBRAS published a dataset to be used in the development of Abnormal Event Management (AEM) (RENGASWAMYB et al., 2003). The dataset, called 3W, was initially published by VARGAS et al.(2019) and republished by PETROBRAS (2022). The 3W dataset presents challenges for the development of a monitoring system. This study aims to explore the detection of failures in oil well operations using the 3W dataset.