Machine Learning-Based Risk Model for Pipeline Integrity Management

The United States maintains about 2 million miles of natural gas pipelines, among which 321,000 miles are oil and gas transmission and gathering pipelines. Most cast-iron pipelines have been in service for more than 100 years. Pipeline failure caused by aging, corrosion, cracks, and damages may result in irreparable societal, economic, and environmental consequences. In this paper, we develop a machine learning-based risk model to predict the failure of pipelines. We train and test the risk model based on the historical data from a report covering 50 years of spillage data in European cross-country oil pipelines. Various factors such as service type, pipeline age, and leak detection are considered in the model. The machine learning-based risk model will provide the pipeline operators with an efficient and effective way to predict the pipeline failure type and guide the pipeline rehabilitation and maintenance procedures. The risk model can also be used to support risk assessment based on the ASME code for pipeline integrity management, help extend the service life of the pipeline, and bring benefits to the environment and the economy.