Abstract
Cementing is an extremely important step in the well construction process. It has important objectives such as hydraulic sealing to prevent migration of undesired fluids from the formations and their collapse. One of the methods to verify the quality of cementat jobs is running acoustic logging tools such as CBL/VDL and ultrasonic and inferring zonal isolation by the interpretation of such data. This study aims to use machine learning techniques for automatic cement logs interpration. Cement logs of 25 wells were used as database. The logs responses have been classified in five classes according to the bond quality by specialized interpreters. These classified segments were used to train neural networks and other supervised machine learning models, such as random forests and k-nearest neighbor (KNN). Feature engineering is used in order to find new and high-performance features. The models were developed in a Jupyter environment using Python libraries. The best classifier has a simple accuracy of 61.4% and approximate accuracy (where the prediction is up to one class away from target) of 91.3%.
