Supervised Machine Learning Applied to Cement Integrity Assessment – A Comparison Between Models and Feature Extraction Techniques

Summary The analysis of the interpretation of the integrity of the bond of the cement layer between the casing and the formation in oil wells has grown significantly as plugging and Abandonment operations also grow. Interpreting this analysis is important to ensure that the well is hydraulically isolated from the surrounding environment before permanently sealing the well. However, this interpretation depends on a specialist's ability to analyze a large demand of data, which is complex and acquired through logging operations. This fact makes this process prone to errors, human dependence, and time-consuming. These three challenges motivated Ouronova, in partnership with Repsol Sinopec Brazil, to develop software to help interpret acoustic profiling. Also inspired by the objective of optimizing the Plugging and Abandonment operations completely, the software also optimizes the planning and management of Plugging and Abandonment campaigns. The so-called plug and abandonment (P&A) Assistant software has proven to be a good tool that optimizes the Plugging and Abandonment process, using machine learning (ML) techniques to improve the accuracy and reliability and reduce the time for cement sheath integrity analysis. This paper then presents some supervised method techniques implemented in the software, such as Logistic Regression, k-Nearest Neighbors, Decision Trees, Random Forest, and Gaussian Naive Bayes. The results show that combining features derived from Cement Bond Log and Acoustic Impedance Log enables the construction of efficient models. The Gaussian model was the one with the best overall performance, achieving a Balanced Precise Accuracy equal to 0.50 and a Balanced Adjacency Accuracy around 0.88.