Master of Engineering Science
Civil and Environmental Engineering
The stochastic process-based models are developed to characterize the generation and growth of metal-loss corrosion defects on oil and gas steel pipelines. The generation of corrosion defects over time is characterized by the non-homogenous Poisson process, and the growth of depths of individual defects is modeled by the non-homogenous gamma process (NHGP). The defect generation and growth models are formulated in a hierarchical Bayesian framework, whereby the parameters of the models are evaluated from the in-line inspection (ILI) data through the Bayesian updating by accounting for the probability of detection (POD) and measurement errors associated with the ILI data. The Markov Chain Monte Carlo (MCMC) simulation in conjunction with the data augmentation (DA) technique is employed to carry out the Bayesian updating. Numerical examples that involve both the simulated and actual ILI data are used to validate the proposed Bayesian formulation and illustrate the application of the methodology.
A simple Monte Carlo simulation-based methodology is further developed to evaluate the time-dependent system reliability of corroding pipelines in terms of three distinctive failure modes, namely small leak, large leak and rupture, by incorporating the corrosion models evaluated from the Bayesian updating methodology. An example that involves three sets of ILI data for a pipe joint in a natural gas pipeline located in Alberta is used to illustrate the proposed methodology. The results of the reliability analysis indicate that ignoring generation of new defects in the reliability analysis leads to underestimations of the probabilities of small leak, large leak and rupture. The generation of new defects has the largest impact on the probability of small leak.
Qin, Hao, "Probabilistic Modeling and Bayesian Inference of Metal-Loss Corrosion with Application in Reliability Analysis for Energy Pipelines" (2014). Electronic Thesis and Dissertation Repository. 2246.