Integrity assessment of corroded oil and gas pipelines using machine learning: A systematic review

Abstract Hydrocarbon fluid integrity evaluation in oil and gas pipelines is important for anticipating HSE measures. Ignoring corrosion is unavoidable and may have severe personal, economic, and environmental consequences. To anticipate corrosion's unexpected behavior, most research relies on deterministic and probabilistic models. However, machine learning-based approaches are better suited to the complex and extensive nature of degraded oil and gas pipelines. Also, using machine learning to assess integrity is a new study field. As a result, the literature lacks a comprehensive evaluation of current research issues. This study's goal is to evaluate the current state of machine learning (methods, variables, and datasets) and propose future directions for practitioners and academics. Currently, machine learning techniques are favored for predicting the integrity of damaged oil and gas pipelines. ANN, SVM, and hybrid models outperform due to the combined strength of the constituent models. Given the benefits of both, most popular machine learning researchers favor hybrid models over standalone models. We found that most current research utilizes field data, simulation data, and experimental data, with field data being the most often used. Temperature, pH, pressure, and velocity are input characteristics that have been included in most studies, demonstrating their importance in corroded oil and gas pipeline integrity assessment. This study also identified research gaps and shortcomings such as data availability, accuracy, and validation. Finally, some future suggestions and recommendations are proposed.