An Approach For Acquiring Knowledge In Complex Domains Involving Different Data Sources And Uncertinty In Label Information: A Case Study On Cementation Quality Evaluation

Oil and Gas area presents many problems in which the experts need to analyze different data sources and they must be very specialized in the domain to correctly analyze the case. So, approaches that uses artificial intelligence techniques to help the experts to help them turning explicit their expert knowledge and analysing the cases is very important. Analysing cementation quality in oil wells is one of these cases. Primary cementation operation of an oil well is creating a hydraulic seal in the annular space formed between the coating pipe and the open well wall, preventing the flow between different geological zones bearing water or hydrocarbons. To evaluate the quality of this seal at determined depths, acoustic tools are used, aiming to collect sonic and ultrasonic signals. Verifying the quality of the available data for cementation quality evaluation is a task that consumes time and effort of the domain experts, mainly due to data dispersion in different data sources and missing labels in data. This work presents an approach for helping acquiring knowledge from domains where these problems are presented using machine learning. Interactive labeling and multiple data sources for acquiring knowledge from experts can help to construct better systems in complex scenarios, such as cementation quality. We obtained promising results in our case study scenario.