Perpustakaan Digital ITB

Well cementing is critical in maintaining the integrity of oil and gas wells during drilling and production. However, deviations from the expected outcomes can occur, highlighting the need to evaluate the bond between cement, casing, and formation after cementing. Acoustic logging is commonly used for this evaluation, but its limitations can lead to misinterpretations of the acoustic signals. As an alternative, ultrasonic logging is employed due to its perceived advantages over acoustic logs. Manual cement bond evaluations, performed by trained professionals, are currently time-consuming, complex, and subjective. Different interpretations can arise even when analyzing the same data, underscoring the need for a more consistent and efficient approach. Machine learning are introduced to automate the interpretation process to address this. By developing a machine learning-based software, the study aims to assist cement bond evaluations. This software generates quick-look interpretations, which serve as a reference for professionals conducting manual interpretations. The outcomes will undergo analysis, and subsequent actions will be determined based on the findings. With Extra Trees as the best algorithm that has validation scores of 93%, the software could predict one casing or liner section cement bond in less than 1 minute. It improves the speed and consistency of cement bond evaluations.