Automated extraction and interpretation of data from cement bond logs legacy documents

Cement placement is a critical component of the well architecture for ensuring mechanical support of the casing, protection from fluid corrosion, isolation of permeable zones at different pressure regimes to prevent hydraulic communication, and effective plug-and-abandonment operations. Companies typically evaluate cement by running cement bond logs (CBL), and/or VDL, that displays the full waveform.

Therefore, the interpretation of these logs provides a clear view of the quality of cement bonds with casing and formation. However, oftentimes, these logs are available as raster images and need to be digitized before interpreting them.

Another important issue in CBL’s is that their interpretation is made manually and under pressure, it can be complex, time consuming, and quite subjective as well.

We developed and tested a web-based solution, which transforms CBL images into sets of values and interprets them in terms of cement quality in a semi-automatic fashion. Our aim was twofold: on one hand, to develop an accurate workflow that allows to save a considerable amount of time with respect to the existing software, while ensuring a very high accuracy. On the other hand, the web solution has been developed to provide the user with an easy to use and fast tool for interpreting the cement quality in boreholes, instead of making it manually.

In the case of black-and-white raster image logs, we used a manual parametrization gridlines-elimination, that intervenes on post-processed data. This provided good results in terms of accuracy and efficiency and has been already implemented in the web solution. Advanced deep learning techniques based on image segmentation seem to give promising results and pave the way for further R&D work in the future.