Doctor of Philosophy
Civil and Environmental Engineering
Cracking is a well-known failure mechanism that threatens the structural integrity of energy pipelines. As a special type of cracking, stress corrosion cracking (SCC) occurs if suitable tensile stress and corrosive environment are present simultaneously. When the basicity of the local electrolyte is close to 7, the failure mechanism is termed as near-neutral pH stress corrosion cracking (NNpHSCC). Cracks, including NNpHSCC, markedly diminish the burst capacity of pipelines through reducing their local wall thickness. Although commonly observed on in-service pipelines and being one of the leading causes of pipeline failures, the studies on cracks, especially NNpHSCC, have not yet received sufficient attention in academia. This thesis conducts a general integrity assessment of pipelines containing cracks and NNpHSCC defects from different aspects using various research tools and methodologies.
The first study presents a review of four existing growth models for NNpHSCC defects on buried oil and gas pipelines: Chen et al.’s model, two models developed at the Southwest Research Institute (SwRI) and Xing et al.’s model. The predictive accuracy of these growth models is investigated based on crack growth rates obtained from full-scale tests conducted at the CanmetMATERIALS of Natural Resources Canada using pipe specimens that are in contact with NNpH soils and subjected to cyclic internal pressures. The comparison of the observed and predicted crack growth rates indicates that the hydrogen-enhanced decohesion (HEDE) component of Xing et al.’s model leads to on average reasonably accurate predictions. The predictive accuracies of the other three models are markedly poorer.
The second study applies the mechanics-based approach and five machine learning (ML) algorithms to classify the failure mode (leak or rupture) of steel oil and gas pipelines containing longitudinally oriented surface cracks. The employed ML algorithms consist of three single learning algorithms, and two ensemble learning algorithms. The classification accuracy of the mechanics-based approach and ML algorithms are evaluated based on full-scale burst tests of pipe specimens collected from the open literature. The analysis results reveal that the mechanics-based approach leads to highly biased classifications: many leaks erroneously classified as ruptures. In contrast, ML algorithms lead to markedly improved accuracy, and the ensemble learning algorithms yield superior classification performance compared to the single learning algorithms. The rationale behind these observations is also thoroughly discussed.
The third study presents the improvement of a widely used burst capacity model for steel oil and gas pipelines that contain longitudinal external surface cracks, namely the CorLAS model, through the addition of a correction factor that is quantified by the Gaussian process regression (GPR). The correction factor is assumed to depend on four non-dimensional input features that characterize both the crack geometry and pipe material properties. A database consisting of full-scale burst tests of pipe specimens that contain longitudinal surface cracks is established based on the open literature, which is employed to train the GPR model and evaluate its performance. It is shown that GPR is highly effective in improving the accuracy of the CorLAS model predictions. The improvement is further shown to have a marked effect on the time-dependent probability of burst of pipelines containing growing surface cracks.
The fourth study conducts time-dependent system reliability analysis of pipelines containing multiple longitudinal surface cracks considering leak and rupture. The Gaussian process-based ML algorithms are harnessed for multiple purposes, encompassing the determination of burst capacity (this endeavor has been successfully accomplished within the scope of the third study), the formulation of a model for segregating the two failure modes, and the creation of surrogate models for two distinct NNpHSCC growth models. The impacts of the spatial variability of various pipe attributes, material properties and environmental conditions on the system reliability are investigated. The Gaussian process-based ML algorithms are shown to be highly effective in identifying the failure modes and predicting the crack growth. The system reliability analysis results indicate that the probability of leak increases more rapidly than the probability of rupture as time increases. Moreover, the spatial variability of the majority of the random variables considered in this study has only marginal effects on the system failure probability.
Summary for Lay Audience
Pipeline transportation is the most cost-effective and efficient means to deliver large volumes of fuels, such as crude oil, natural gas, carbon dioxide and hydrogen, over long distances, and is a critical part of the energy infrastructure in a modern society. Various failure mechanisms (e.g. metal-loss corrosion, stress corrosion cracking, third-party interference, ground movement, etc.) pose threats to the structural integrity of buried steel pipelines, mainly through reducing the pressure containment capacity, i.e. burst capacity, of defected pipelines, leading to potential failures. Cracking is widely considered as a critical failure mechanism as it could cause sudden failure with no prior warning. Crack propagations on pipelines can be accelerated in corrosive environments whose pH are close to 7. If tensile stresses are simultaneously present, such cracks are termed as near-neutral pH stress corrosion cracking (NNpHSCC) defects, which is one of the leading causes of pipeline failures. This thesis does some work on the integrity management practice of pipelines containing generic cracks and NNpHSCC defects.
Some empirical and semi-empirical growth models for NNpHSCC have been proposed in the literature. This thesis reviews these models and assesses their predictive accuracy using data obtained from a full-scale NNpHSCC growth test program. As for generic cracks, it is noticed that the traditional mechanics-based approach to separate two failure modes of cracked pipelines, i.e. leak and rupture, is highly biased, and the industry-adopted CorLAS model to predict the burst capacity of cracked pipelines is associated with considerable model uncertainty. This thesis employs different machine learning algorithms to categorize the two failure modes more accurately, and to improve the predictive performance of the CorLAS model, using full-scale burst test data. The implication of the improvement for the reliability analysis is also investigated. Given these applications of machine learning, a time-dependent system reliability analysis of pipelines containing multiple NNpHSCC defects is conducted, considering multiple failure modes.
Sun, Haotian, "Methodologies for the Integrity Assessment of Pipelines Containing Cracks" (2023). Electronic Thesis and Dissertation Repository. 9463.
Available for download on Saturday, August 31, 2024