AI-Based Cement Bond Quality Assessment: A First Step for Optimizing P&A Design

Abstract

Summary As decommissioning operations continue to expand, the challenges associated with evaluating the integrity of the cement layer and its bond to casing and formation become more pronounced. Ensuring hydraulic isolation of the wellbore from the surrounding environment is crucial before permanently sealing the well. However, the current methodology relies on the expertise of specialists who interpret extensive and intricate data obtained through logging operations. Recognizing the challenges inherent in cement evaluation, Ouronova, in collaboration with Repsol Sinopec Brazil, is developing a computational solution to help specialists interpret cement integrity logging data. Simultaneously, the developed tool aims to assist operators in optimizing the planning and management of decommissioning campaigns. The innovative software employs machine learning techniques that have exhibited significant promise in enhancing accuracy, reliability, and efficiency in the analysis of cement sheath integrity. Thus, the objective of this paper is to present some results obtained with the software by using Convolutional Neural Networks to predict the cement condition in two wellbore regions. The acquired dataset was used to generate Variable Density Logs diagram and plots here referred to as 2D Combined Signals, which were used as inputs to train the model. The main results indicate good accuracy in predicting the cement condition using the Variable Density Log and the 2D Combined Signals. In special, the latter showed to be a more promising option because its accuracy value tended to be more stable as the database was increased, in comparison with the Variable Density Log case. As a metric for the comparisons, the Balanced Adjacency Accuracy was used. For the results based on the Variable Density Log, we found a value of 0.810, while for the ones based on the 2D Combined Signals, we found 0.958.