Abstract
One of the key elements of well abandonment process is the restoration of regional cap rocks, typically achieved by understanding of quality of cement behind casing by means of interpretation of cement bond logs. The main challenges of cement evaluation, however, are the availability of these logs where these are required, the cost of the acquisition of the data, and well accessibility issues which sometimes prevent the acquisition of the data. The intent of the project is to use machine learning technology to provide an assessment of cement quality based on learning datasets as inputs. Presented here is Phase 1 of a longer study as a proof of concept. Relevant data in the form of petrophysical data, geomechanics data and cement job reports were collated, and then machine learning techniques were applied to identify key parameters which may impact prediction of quality of cement, and subsequently algorithms can then be generated.
