Automated Corrosion Log Quality Control and Interpretation Using Machine-Learning

Abstract

With the increasing focus on data mining and machine learning (ML) applications in the oil and gas industry, the substantial number of well integrity logs and variety of data types represent a suitable candidate for the implementation of automated corrosion log processing. Convolutional neural networks (CNNs) are used in many fields, especially for image processing and features recognition. On the other hand, genetic algorithms (GA) add a valuable benefit to data processing in terms of global search and optimization. This paper demonstrates the integration of ML techniques with legacy well integrity log data, improving the results and leading to a tangible time and cost savings. A downhole well integrity evaluation triangle comprises three important services for comprehensive diagnosis: 1) cement evaluation, 2) corrosion inspection, and 3) leak detection. These services produce multiple datasets from a variety of logging tools. The types and sources of these datasets include synthetic data from simulation and modeling, tool calibration and lab testing, as well as raw, processed, and interpreted data. This paper describes the use of advanced ML techniques to scrutinize and improve well integrity evaluation. The new process resolves the recurrent challenges of well integrity evaluation in complex completion and downhole environments. It also maximizes value from existing well and field data. Image features recognition enables major improvements in the data analysis, such as the identification of concentric casings and tubing as well as their respective collar depths and types. In addition, input parameters and well schematics promote quality control of recorded data versus the model data. The new process helps to identify casing and completion accessories and provides a reliable benchmark. Another major element is the qualitative and quantitative evaluation of corrosion using deep learning algorithms combined with the GA. This evaluation is achieved using feature extraction from the forward model (FM) data in an analogous way to collar identification in an electromagnetic decay image. The integration of big data and advanced ML enables an improved data analysis with automated data quality control (QC) and interpretation. A more pro-active well integrity management system will result. Testing and validation using field examples demonstrate the benefits of this new methodology. The outcome is a better-quality answer product that helps depict various aspects of the acquired and interpreted data. Savings in time and cost are complemented with an improved and automated quality control.