Perpustakaan Digital ITB

There are so many mature wells that do not produce any oil or gas anymore in our current time. This well becomes more mature and only becomes a heritage for the next generation. However, there is a solution to reactivate this well to become productive again. This solution needs some consideration, including the cement bond strength. Wells that have poor cement bonds will have an opportunity to cause production leaking. Leaking in production activity will affect the well quality and increase other dangerous problems, for example, blow-out. That is the reason to check the quality of the cement bond before reactivating the well. In order to find the cement bond quality, expertise usually runs a cement bond log. This logging tool will give data, including transit time, amplitude, variable intensity, and oscilloscope pictures. These 4 data will be used by expertise to interpret the cement bond quality and distinguish it into two parts: good cement and poor cement bond. However, one expert will have a different interpretation from the other expertise. It happens because of the diverse experience that every expertise has. Even though this difference happens, in general, every expertise has a basis that they use to interpret the cement bond log data. By using machine learning interpretation in the early interpretation time, general understanding can be done faster, and expertise will have more time to discuss the differences. This study makes an application to interpret the data that is not interpreted yet by learning expertise habits. The learning model is acquired using six different well log data with 47,789 data along with more than seven kilometers log that has been manually interpreted based on cement bond log amplitude and gamma-ray. The machine learning model is distinguishing into five models. Every model gives a different interpretation. However, the results still show that the model has an accuracy of more than 75%.