Bayesian updating of stochastic process-based models for corroding gas pipelines based on imperfect inspection information

This paper presents an efficient approach for Bayesian analysis of corroding gas pipelines containing metal-loss corrosion defects subjected to internal pressure. The methodology considers stochastic dependence among individual defects. The Nataf model is used to model the interdependence (correlation) among the defects. The generation of new defects and the growth of existing ones are included in the stochastic modelling, by employing a Non-Homogeneous Poisson Process (NHPP) and a Homogeneous Gamma Process (HGP), respectively. Information on a corroding pipeline obtained through multiple In-Line Inspections (ILIs) is used for the Bayesian updating of stochastic models. The Probability of Detection (PoD) and measurement error of the inspection tools are also taken into consideration. The Bayesian updating is conducted by using Structural Reliability Methods (SRM), a recently proposed method in literature that sets an analogy between Bayesian updating and a reliability problem. The SRM adopted herein is Subset Simulation (SuS) and the whole analysis is referred to as BUS-SuS. The updating is conducted in conjunction with the Data Augmentation (DA) technique which accounts for both detected and undetected defects, by treating the undetected ones as the missing data. Multiple simulated corrosion defects from different ILI inspections are employed for the implementation and validation of the methodology. A parametric study that examines the impact of correlations among defects on the posterior stochastic models is also included in the numerical example.