Applications of machine learning in pipeline integrity management: A state-of-the-art review

Abstract Despite being considered the safest means to transport oil and gas, pipelines are susceptible to degradation. Pipeline integrity management (PIM) is implemented to lower the risk of failure due to degradation and to maintain the functionality and safety of pipelines. PIM consists of a set of activities for assessing the operational conditions of pipelines. These activities generate data with high volume, velocity, and variety, due to the length of a pipeline and the number of sensors and tools used to assess the pipeline's condition. This paper provides a comprehensive review in relation to the applications of machine learning (ML) in managing and processing data generated from PIM activities. ML applications in the elements of a PIM process (e.g., inspection, monitoring, and maintenance) are investigated. The aspects of ML techniques (i.e., type of input, pre-processing, learning algorithm, output and evaluation metric) applied in each element of PIM are examined. Current research challenges and future research opportunities in the application of ML in PIM are also discussed.