Machine learning framework to generate synthetic cement evaluation logs for wellbore integrity analysis

Proper zonal isolation is key to ensure optimum injection and production and it is highly dependent on the cement bond to the casing and the formation. Yet, in most wells, cement evaluation logs are not run. This work will describe a workflow to address the lack of cement evaluation data by creating synthetic logs to aid wellbore integrity analysis. Synthetic logs can be a reliable and cost-effective alternative to predicting the cement bond than running log tools in every well during a drilling campaign, and, with further development, has the potential to be used for well design. A machine learning framework based on Gaussian process regression (GPR) was chosen in the development of this procedure because it can assess uncertainty of the cement bond through estimation of error and confidence interval. GPR also require less training samples than conventional machine learning techniques. In this work CBL data was used for training and the model was validated through comparison with data from a different well in the same field. The results shows that the predicted case correlated very well with the base case, with some curves overlaying even in the poor bond sections. Initial assumptions given by the covariance function help capture not only the general trend relationship but also localized variations, which play a major role in the way a fracture propagates in the annulus. Additionally, the uncertainty assessment provided by this framework can assist risk management by determining worst case scenarios and potential fluid migration paths in the annulus.