Application of wavelet analysis and machine learning on vibration data from gas pipelines for structural health monitoring

Abstract

Abstract Defects due to corrosion and fatigue in pipelines may create significant hazards in the transportation of natural gas. To detect such damages, traditionally, pigging equipment like pipeline inspection gauge (PIG) are inserted into the pipeline. However, the performance of such devices may be enhanced by embedding vibration sensors externally. Such sensors equipped with machine learning capability could be used for damage detection and verification of the PIG data. This study investigates the applicability of Support Vector Machine (a supervised machine learning classifier) and wavelet analysis on vibration response in the detection of various kinds of defects present in gas pipelines. Both ends fixed and simply supported boundary conditions are incorporated on a pipe with outer diameter 200mm pipe with 25mm pipe thickness, made of structural steel to simulate the real transportation pipeline laying conditions above the ground. In this study, the pigging process is simulated in ANSYS by considering inspection gauge as moving load inside the pipeline with a constant velocity. The velocity and acceleration time history data at a fixed point on the pipe for gauge moving from one end to the other is obtained from ANSYS corresponding to different loading conditions and load moving velocities of the inspection gauge. These data are then post-processed in MATLAB environment. Wavelet analysis has been carried out on this data, to obtain spectral components of frequency contained in the data. Further, Support Vector Machine classifier is used to separate the segment of data corresponding to defect region, which can be mapped back to identify the physical location of the defect on the pipeline. The obtained results show good accuracy of defect identification and its location prediction, which can be integrated with intelligent PIG devices and pipe crawling robots.