A Novel Corrosion Monitoring and Prediction System Utilizing Advanced Artificial Intelligence

Abstract

Corrosion in pipes is a major challenge for the oil and gas industry as the metal loss of the pipe, as well as solid buildup in the pipe, may lead to an impediment of flow assurance or may lead to hindering well performance. Therefore, managing well integrity by stringent monitoring and predicting corrosion of the well is quintessential for maximizing the productive life of the wells and minimizing the risk of well control issues, which subsequently minimizing cost related to corrosion log allocation and workovers. We present a novel supervised learning method for a corrosion monitoring and prediction system in real time. The system analyzes in real time various parameters of major causes of corrosion such as salt water, hydrogen sulfide, CO2, well age, fluid rate, metal losses, and other parameters. The data are preprocessed with a filter to remove outliers and inconsistencies in the data. The filter cross-correlates the various parameters to determine the input weights for the deep learning classification techniques. The wells are classified in terms of their need for a workover, then by the framework based on the data, utilizing a two-dimensional segmentation approach for the severity as well as risk for each well. The framework was trialed on a probabilistically determined large dataset of a group of wells with an assumed metal loss. The framework was first trained on the training dataset, and then subsequently evaluated on a different test well set. The training results were robust with a strong ability to estimate metal losses and corrosion classification. Segmentation on the test wells outlined strong segmentation capabilities, while facing challenges in the segmentation when the quantified risk for a well is medium. The novel framework presents a data-driven approach to the fast and efficient characterization of wells as potential candidates for corrosion logs and workover. The framework can be easily expanded with new well data for improving classification.