Protecting steel pipeline systems from failure due to corrosions defects is a crucial issue in pipeline industry. Reliability models that use the rate of corrosion growth combined with closed form solutions for the failure pressure are often used to estimate the time periods before excavation and repair. A methodology is presented for the assessment of predicted failure pressure based on finite element analysis (FEA) and reliability analysis. Deterministic failure equations are transformed to probabilistic limit state models. The failure mode is considered to be controlled by the stresses due to internal pressure and the presence of corrosion. A response surface method (RSM) is utilized to build a surrogate model of the limit state function. A comparison between closed-form and the surrogate model approach is discussed. A stochastic model is assumed to match the uncertainty inherent in both loads and strength. Simulation-based approaches and asymptotic methods for probability of failure evaluation are used, namely, Monte Carlo simulation, importance sampling, First Order Reliability Method (FORM) and Second Order Reliability Method (SORM). An adaptive building of the numerical experimental design for the surrogate limit state is proposed. A new artificial neural network (ANN) is developed in order to reduce the computational cost of experimental design scheme’s evaluation. The outcomes obtained from such an approach are useful as a decision-making tool for the maintenance, repair or optimization of pipelines systems.

This content is only available via PDF.
You do not currently have access to this content.