Prediction of oil and gas pipeline failures through machine learning approaches: A systematic review

• Main Authors: Al-Sabaeei, A.M., Alhussian, H., Abdulkadir, S.J., Jagadeesh, A.
• Format: Article
• Published: Elsevier Ltd 2023
• Online Access: http://scholars.utp.edu.my/id/eprint/37299/; https://www.scopus.com/inward/record.uri?eid=2-s2.0-85168241856&doi=10.1016%2fj.egyr.2023.08.009&partnerID=40&md5=c11a628b554bd74ac80009361e25c404

Pipelines are vital for transporting oil and gas, but leaks can have serious consequences such as fires, injuries, pollution, and property damage. Therefore, preserving pipeline integrity is crucial for a safe and sustainable energy supply. The rapid progress of machine learning (ML) technologies provides an advantageous opportunity to develop predictive models that can effectively tackle these challenges. This review article mainly focuses on the novelty of using machine and deep learning techniques, specifically artificial neural networks (ANNs), support vector machines (SVMs) and hybrid machine learning (HML) algorithms, for predicting different pipeline failures in the oil and gas industry. In contrast to existing noncomprehensive reviews on pipeline defects, this article explicitly addresses the application of ML techniques, parameters, and data reliability for this purpose. The article surveys research in this specific area, offering a coherent discussion and identifying the motivations and challenges associated with using ML for predicting different types of defects in pipelines. This review also includes a bibliometric analysis of the literature, highlighting common ML techniques, investigated failures, and experimental tests. It also provides in-depth details, summarized in tables, on different failure types, commonly used ML algorithms, and data resources, with critical discussions. Based on a comprehensive review aforementioned, it was found that ML approaches, specifically ANNs and SVMs, can accurately predict oil and gas pipeline failures compared to conventional methods. However, it is highly recommended to combine multiple ML algorithms to enhance accuracy and prediction time further. Comparing ML predictive models based on field, experimental, and simulation data for various pipeline failures can establish reliable and cost-effective monitoring systems for the entire pipeline network. This systematic review is expected to aid in understanding the existing research gaps and provide options for other researchers interested in predicting oil and gas pipeline failures. © 2023 The Author(s)