Assessment of the structural conditions in steel pipeline under various operational conditions – A machine learning approach

•The monitoring of leakage in pipelines is performed using a network of FBG sensors.•Strain data are used to detect leakage under different operational conditions.•SVM algorithms are implemented to predict the structural conditions of the pipe.•The leakage detection and localization accuracy obtained is above 95%. Oil and water transport pipeline systems are susceptible to damage due to harsh environmental conditions and operational factors. Hence, ongoing maintenance and inspection are required. The development of continuous and reliable monitoring will ensure the safe usage of these structures and assist in the extension of their life spans. In this study, the monitoring and assessment of pipelines are performed using a network of Fiber Bragg Grating (FBG) sensors mounted in the longitudinal and circumferential directions pipelines. The sensitivity of these measurements to assess pipe pressure and flow variations, and leakage detection and localization were evaluated. Water, at a controlled pressure and flow rate, was pumped into the designed 6-m pipe testbed. Leakage was simulated by opening one of the four designated valves installed on the pipe. Support Vector Machine (SVM) algorithms were implemented, using the collected data, to assist in the prediction of the structural condition of the pipe under various operational conditions. Pressure variations inside the pipe highly impacted the amplitude of the measured strain, increasing it significantly up to 20%. A flow rate increase of 5 GPM had the inverse effect, resulting in a 5% decrease in the amplitude of the measured strain. A change of leakage hole size greatly influenced the measured signal, resulting in a 55% change in amplitude between a 2-cm2 and a 5-cm2 hole. To determine the leakage location, only the temporal aspects of the signal were affected, resulting in slight shifts in sensor response times. The developed SVM classifiers reached accuracies of 88% for flow rate classification, greater than 95% for pressure classification, and 100% for leakage size classification. The accuracy of leakage localization did not exceed 72%. These results are promising for the monitoring of the structural conditions related to leakage detection and localization, based on the patterns observed.