Pipeline Rupture Detection Based on Machine Learning and Pattern Recognition
Pipeline ruptures have the potential to cause significant economic and environmental impact in a short period of time, therefore it is critical for pipeline operators to be able to promptly detect and respond to them. Public stakeholder expectations are high and an evolving expectation is that the response to such events be automated by initiating an automatic pipeline shutdown upon receipt of rupture alarm. These types of performance expectations are challenging to achieve with conventional, model-based, leak-detection systems (i.e. CPM–RTTMs) as the reliability measured in terms of the false alarm rate is typically too low. The company has actively participated on a pipeline-industry task force chaired by the API Cybernetics committee, focused on the development of best practices in the area of Rupture Recognition and Response. After API’s release of the first version of a Rupture Recognition and Response guidance document in 2014, the company has initiated development of its own internal Rupture Recognition Program (RRP). The RRP considers several rupture recognition approaches simultaneously, ranging from improvements to existing CPM leak detection to the development of new SCADA based rupture detection system (RDS). This paper will provide an overview of a specific approach to rupture detection based on the use of machine learning and pattern recognition techniques applied to SCADA data.