Magnetic flux leakage data analysis for oil and gas pipeline integrity management

Corrosion is a significant cause of oil and gas pipeline failures. In the pipeline industry, in-line inspection (ILI) using magnetic flux leakage (MFL) technique is conducted periodically to detect and assess pipeline corrosion. The ILI data is a critical input of the pipeline integrity management (PIM) program, where the decision on the pipeline integrity maintenance is made. This thesis research aims to facilitate the decision-making process in the PIM program from the perspective of MFL data analysis. First, the concept of parameterization is put forward to obtain a contextual representation of the corrosion defect. In the PIM program, this high-level representation can help structural engineers to retrieve similar corrosion defects that could pose serious threats to the pipeline integrity. Three parameterization models, i.e., principal component analysis, convolutional auto-encoder, and shape context, are proposed to achieve the contextual defect representation. Then, a computational framework is proposed to automatically match the multiple inspection results from different tools, i.e., axial MFL and circumferential MFL. Due to their complementary detection capabilities, the matched multi-modal MFL data can be further integrated to obtain more comprehensive defect assessment. The proposed framework employs a sliding window searching approach and a Gaussian mixture model to align the coordinate systems of two data sets. In the aligned coordinate system, an accurate matching is achieved with a modified density-based spatial clustering of applications with noise algorithm, which considers both the location and the size information of the corrosion defect. Last but not least, the detection performance of MFL inspection is assessed quantitatively. It aims to figure out which defect can be reliably detected and which may be missed in the MFL inspection. Therefore, even the undetected defects could also be considered in the PIM program. A probability of detection (POD) model is proposed to realize the quantitative performance assessment. Due to the characteristics of MFL inspection, the proposed POD model is constructed with two geometric variables, i.e., the volume and the orientation of a corrosion defect. Besides, inspection results from different tools are integrated to study the POD of their combination. The research outcomes in this thesis contribute to the PIM program from different perspectives. Contextual defect representation employs the MFL data from individual inspection to identify the pipeline integrity threat. MFL data matching is the precondition to integrate the inspection results from different MFL tools and eventually obtain a comprehensive corrosion defect assessment. Besides, the detection performance assessment of MFL inspection takes the undetected defects into consideration and ensures no defects are ignored in the PIM program.Applied Science, Faculty ofEngineering, School of (Okanagan)Graduat