Automatic Depth Alignment of High-Resolution Magnetic Flux Leakage Data to Detect Corrosion in Downhole Casing Using Machine Learning

Abstract Periodically monitoring well casing and tubing integrity is essential for making timely decision for well intervention to safely operate oil and gas wells. Magnetic flux leakage (MFL) inspection tools are most used in downhole casing inspections. Due to different positions of the sensors on the inspection tool, the MFL data measured by the sensors are misaligned. These data need to be aligned before they can be used for further analysis. Oil and gas operators rely on human experts to align and analyze these data to assess corrosion defects in casing. This analysis done by different individuals at different times, using different data processing tools supporting their interpretation, often leads to inconsistent risk evaluation. In this paper, we demonstrate results from a machine learning (ML) based approach that automatically aligns multiaxial MFL inspection data, enabling consistent and accurate identification of corrosion defects in well casing. High resolution MFL casing inspection tool used in this study consists of dual inspection modules – upper and lower. Each inspection module houses 48 tri-axial flux leakage (FL) sensors and 48 discriminator (DIS) sensors. Data from casing inspection included – 1) Memory data: high-resolution flux leakage data from FL and DIS sensors, acceleration data stored in tool’s memory, 2) Monitoring data: low-resolution flux leakage data from FL and DIS sensors, acceleration data, winch speed, and tool depth stored in the surface data acquisition system. We applied a pattern recognition technique to identify the casing joint collar signatures in flux leakage data. Then we used Genetic Algorithm (GA) to align collars identified in – 1) upper and lower modules data, 2) memory and monitoring data and 3) monitoring data and previous inspections reports with physical depth of collars. The final output of our algorithm is the aligned high-resolution MFL data from memory, associated with physical depth, suitable for corrosion defect evaluation. We used historical casing inspections data, reports and gamma-ray and neutron correlation logs from a vertical well in North America (Well 1) to demonstrate results from our algorithm. Accuracy of the depth alignment of the high-resolution MFL data was evaluated using the average Euclidean distance (mm) between the casing joint collars in memory, monitoring data and previous inspections reports. Our proposed algorithm outperformed the existing methods by achieving less than 50 mm or 2 inches misalignment between 90% of the collars in the upper and lower modules. It also achieved reasonable alignment between 75% of the collars identified in monitoring and memory data, as well as, between previous reports and in the memory data. MFL casing inspection data are noisy due to vertical and rotational movement of the tool inside the well casing. These data also include missing and erroneous data from sensor failures. The proposed method is robust enough to perform well under such circumstances. Traditional methods rely on human experts to manually align MFL data for detecting corrosion in casing, leading to inconsistent risk evaluation of trends. The main contribution of this paper is the automatic depth alignment of the high-resolution MFL data, by analyzing the historical inspection reports. This enables higher accuracy and consistency in assessing corrosion progression in well casing over time.