Thesis Format
Integrated Article
Degree
Doctor of Philosophy
Program
Civil and Environmental Engineering
Supervisor
Zhou, Wenxing
Abstract
Metal-loss corrosion is one of the primary failure mechanisms that threaten the structural integrity of buried oil and gas steel pipelines. Pinhole is one of the corrosion mechanisms that can reduce the burst capacity of pipelines as a result of localized wall thickness reduction and pose unique challenges to the pipeline corrosion management programs. A pinhole could be associated with a patch of corroded area surrounding the pinhole, resulting in a particular type of complex-shaped defect referred to in this study as a pinhole-in-general corrosion (PIC) defect. High-resolution in-line inspection (ILI) tools are routinely used in the pipeline industry to detect and size corrosion anomalies. As the difficulty of accurately estimating PIC sizes and burst capacity of pipelines containing PIC defects, the use of machine learning (ML) tools provides a viable option to predict the size and location of PIC defects and develop the corresponding burst capacity models with high accuracy. The main objective of the present thesis is to facilitate the inspection and burst capacity assessment of complex defects including pinholes based on ML tools.
The first study employs the convolutional neural network (CNN) to predict the dimensions and locations of corrosion defects on steel pipelines based on magnetic flux leakage (MFL) signals. Extensive three-dimensional parametric finite element analyses are carried out to generate MFL signals corresponding to semi-ellipsoidal-shaped corrosion defects with different sizes and locations on a pipe model. The white noises characterized by different signal-to-noise ratios are considered in the analysis to represent the measurement errors in the real MFL inspection tool. The numerically generated MFL signals are used to train and validate the CNN model to predict the dimensions and locations of the corrosion defects. The results indicate that the developed CNN model achieves a high predictive accuracy. The study demonstrates the application of CNN model to improve the pipeline integrity management practice.
The second study develops two CNN models to respectively identify pinholes and predict the sizes and location of the identified pinhole according to the magnetic flux leakage signals generated using the magneto-static finite element analysis. Extensive three-dimensional parametric finite element analysis cases are generated to train and validate the two CNN models. The proposed classification CNN model is shown to be highly accurate in classifying pinholes and pinhole-in-general corrosion defects. The proposed regression CNN model is shown to be highly accurate in predicting the location of the pinhole and obtain a reasonably high accuracy in estimating the depth and diameter of the pinhole even in the presence of measurement noises. This study indicates the effectiveness of employing deep learning algorithms to enhance the integrity management practice of corroded pipelines.
The third study carries out extensive parametric three-dimensional elasto-plastic finite element analysis (FEA) to evaluate the burst capacity of oil and gas pipelines PIC defects. The analysis results reveal that the burst capacity of the PIC defect is insensitive to the pinhole diameter but largely affected by the depth of the pinhole and length of the general corrosion. A pinhole located at the centre of the general corrosion is found to have a larger impact on the burst capacity than the same pinhole located near the edge of the general corrosion. The observed size and location effects on the burst capacity are attributed to the bulging deformation of a corroded pipeline under internal pressure. The findings of this study provide the basis for developing a practical, accurate engineering burst capacity model for PIC defects.
The fourth study develops a practical burst capacity model for oil and gas pipelines with PIC defects based on results of extensive parametric three-dimensional FEA that are validated by full-scale burst tests of corroded pipe specimens reported in the literature. The PIC defect considered in this study contains a cylindrical-shaped pinhole located inside a cuboidal-shaped general corrosion. The proposed model takes the form of the well-known PCORRC model for the burst capacity of the general corrosion plus a correction term taking into account the impact of the pinhole. The accuracy of the proposed PIC model is validated using 16 FEA cases and shown to be higher than the well-known RSTRENG model and a burst capacity model for complex-shaped corrosion defects reported in recent literature.
Summary for Lay Audience
Pipelines are critical for transporting large quantities of oil and gas efficiently, such as crude oil and carbon dioxide. However, these pipelines can suffer from defects such as corrosions, which threaten their structural integrity and reduce their ability to contain pressure, known as burst capacity. Pinhole corrosions, which are a particular form of corrosion, are often observed on oil and gas pipelines. Due to the characteristics of pinhole corrosion such as small length and width but significant depth, they can pose unique challenges to the pipeline corrosion management program including detection and fitness-for-service assessments. Pinhole-in-corrosion (PIC) is a complex corrosion which lacks of in-depth studies on its inspection and burst capacity prediction. This thesis focuses on using advanced technologies like machine learning and finite element (FE) simulations to improve the detection and burst capacity assessment of this type of defect, ultimately enhancing the safety and reliability of pipelines.
Convolutional neural network (CNN) is employed in this study to predict both defect profile and location of corrosions on steel pipelines based on FE-generated noisy magnetic flux leakage (MFL) signals with high accuracy and efficiency. Building on this, two CNN models are developed specifically for identifying and measuring tiny corrosion spots, known as pinholes. Extensive FE simulations were used to train these models, and the results showed that the CNN could accurately detect and measure pinholes from different types of corrosion defects, including complex defects. This study also investigates how pinhole size and location could impact the burst capacity of pipelines based on large amount of parametric FE cases. Based on the observations, a new burst capacity model is proposed for PIC defects using Gaussian process regression (GPR).
Recommended Citation
Shen, Yufei, "Application of Machine Learning Tools in the Detection, Sizing and Burst Capacity Prediction for Corrosion Defects on Pipelines" (2024). Electronic Thesis and Dissertation Repository. 10429.
https://ir.lib.uwo.ca/etd/10429