Detection of faults in subsea crude oil pipelines by machine learning assisted process monitoring

Eastvedt, Daniel Theodore (2021) Detection of faults in subsea crude oil pipelines by machine learning assisted process monitoring. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (2MB)

Abstract

Crude oil pipelines are a major infrastructure investment for subsea oilfields and for platform-to-shore transportation and continue to be the preferred method of transporting crude oil over short to medium distances due to low operating costs and high safety record compared to other transport methods. Their remote and often extreme locations limit their access and increases risks of human and environmental hazards in the event of failure. Due to their frequent placement in deep water and hostile environments, inspection and identification of faults (such as leaks and flow restrictions) are difficult, expensive, and hazardous. Faults are often only identified in accidents or upon routine inspection and after significant material losses and/or environmental damage has occurred. The oil industry would benefit from a low-cost and timely method of fault detection. This thesis proposes such a method by augmenting process monitoring with Machine Learning (ML). This thesis investigates the relationship between pressure change, velocity change, and temperature of crude oil through a pipeline. A representative dataset of crude oil flow is generated by computational fluid dynamics (CFD) and used to train a ML algorithm to develop a model of fluid behavior under normal pipeline operations over a range of typical flow rates and temperatures. CFD data are then collected under several simulated fault conditions: leaks of 10 and 20% of the inner cross-sectional area of the pipe, and a restriction to flow of 50% of the cross-sectional area. This thesis demonstrates that the ML algorithm can be trained to model the system under normal conditions, thereby successfully recognizing a fault condition as non-conforming and indicative of a statistically significant change in pipeline operation. It is further able to identify the fault type based on the pattern observed in the new data. This work demonstrates that ML may be a low-risk, low-cost, and accurate method of monitoring a subsea crude oil pipeline for optimal performance and fault detection without the need to introduce special equipment to a subsea pipeline network, assuming flowmeters and temperature probes are employed for process monitoring. This thesis develops the model algorithm, and it is hoped that the results of this study provide a basis for the integration of machine learning and further “big data” techniques in loss prevention and health and environmental protection in the offshore oil industry and elsewhere.

Actions (login required)

View Item