Finite Element Analyses and Random Field Modeling of Naturally Corroded Underground Energy Pipelines
Abstract
Metal-loss corrosion is one of the major threats to the integrity of oil and natural gas pipelines. The Fitness-For-Service (FFS) assessment is commonly carried out to demonstrate the integrity of the corroded pipelines. The naturally occurring corrosions on the external surfaces are irregularly shaped three-dimensional features. Due to the geometric complexity of the natural corrosion features, the research of FFS assessment in the previous studies always employed the pipe segments containing artificially induced corrosion features. However, since artificial features are in general regular-shaped, e.g. cubic or semi-ellipsoidal, they do not capture geometric characteristics of naturally occurring corrosions. The thesis develops a finite element model and a random field based corrosion model to deal with five issues regarding the FFS assessment of pipe segments containing naturally occurring corrosion features.
The first study develops a three-dimensional finite element model to simulate the full-scale burst tests of pipe segments containing real corrosion features. The finite element model, as well as the RSTRENG model, is used to study the impact of the depth threshold and five commonly used interaction rules on the burst capacity predictions of naturally corroded pipe segments. This study recommends the optimal depth threshold values and interaction rules to evaluate the burst capacity of corroded pipelines in the practical FFS assessment.
The second study investigates the impact of corrosion anomaly classes on the predictive accuracy of the B31G, B31G-M, Shell92, PCORRC, PCORRC-M, CSA and RSTRENG models, based on 897 corrosion anomalies on 16 naturally corroded pipe specimens removed from in-service pipelines. The 897 corrosion anomalies are classified into seven classes, namely pin hole, axial slotting, axial grooving, circumferential slotting, circumferential grooving, pitting and general corrosion, based on the Pipeline Operators Forum (POF) anomaly classification system. The seven burst capacity models and finite element analyses (FEA) are employed to evaluate the burst capacities of the corrosion anomalies. The accuracies of the burst capacity models are assessed and compared based on the FEA-to-model predicted burst capacity ratios for different classes of anomalies.
The third study proposed a modified RSTRENG (RSTRENG-M) model to evaluate the burst capacity of corroded pipelines by employing the riverbed profile of corrosion features. The riverbed profile consists of representative corrosion depths at a collection of circumferential profiles of the corrosion feature and takes into account the influence on the burst capacity of the maximum corrosion depth and overall metal loss associated with the circumferential profile. Based on full-scale burst tests of 16 naturally corroded pipe specimens and 44 specimens containing artificially induced corrosion features, RSTRENG-M is shown to be more accurate than the RSTRENG model, particularly for corrosion features with complex morphologies, comparable to the Psqr model in terms of accuracy but more advantageous in terms of computational efficiency. An empirical equation is also developed to estimate the representative depth for a given circumferential profile directly from the corresponding maximum depth to facilitate the application of RSTRENG-M in the context of the inline inspection data.
The fourth study proposes a random field model to characterize the corrosion depth on the external surface of buried oil and gas pipelines. The model addresses the intermingling of corroded and corrosion-free areas on the pipe surface by using a latent homogeneous Gaussian random field and a spatial position-dependent threshold associated with the latent Gaussian field. High-resolution corrosion measurement data obtained from corroded pipe segments removed from in-service pipelines are used to estimate parameters of the proposed model, including the probability of corrosion at a given point, marginal distribution of the nonzero corrosion depth and correlation structure of the latent Gaussian field. A comparison of simulated and measured corrosion fields suggests that the proposed model is able to capture the characteristics of naturally-occurring corrosion field on the pipe surface.
The fifth study combines the FEA model developed in the second study with the random field-based corrosion model presented in the fourth study to analyze the simulated naturally occurring corrosion features (i.e. synthetic corrosion features) in large quantity to further validate the RSTRENG-M model. For comparison, the predictive accuracies of RSTRENG, CSA, CPS, Psqr and DNV are also investigated based on the FEA results of the synthetic corrosion features. The impact of length and maximum depth of corrosion feature on the predictive accuracies of the six semi-empirical models are also presented.