Volume 19, No. 4, 2022
Application Of Machine Learning Techniques For Leak Detection In Horizontal Pipelines
July Andrea. Gomez Camperos
Abstract
Pipelines are considered as the safest way to transport oil and gas, critical pipeline failures such as leaks affect the reliability of fluid transport systems causing environmental damage, economic losses and pressure reduction in the pipeline, so this paper presents a methodology to detect leaks in pipelines by using machine learning techniques that, by introducing process data from the experimental pipeline, systematically determine whether or not there are leaks. For this research, two machine learning classification techniques, support vector machines and decision trees, were evaluated and four sensors, two for flow and two for pressure, were used in a ½ inch diameter horizontal experimental pipeline installed in the automation and control laboratory of the Universidad Francisco de Paula Santander Ocaña. With the experimental data, a complete database was created and used for training and validation of each of the machine learning techniques used. As a result, a leak detection method was obtained, using a data set for training, validation and testing and with accuracy levels higher than 97% in leak detection.
Pages: 653-663
Keywords: Pipelines are used to transport fluids over long distances and are typically found in water distribution systems and in the petrochemical industry.