Abstract
Bayesian networks (BN) are useful tools for corrosion modeling of complex systems such as oil and gas pipelines. We have developed BN models for assessing risk of pipeline failures due to internal corrosion, external corrosion and third party damages. These models take into account dependencies among all variables and can handle situations with limited/incomplete data. This paper is a case study demonstrating how to perform Internal Corrosion Direct Assessment (ICDA) using BN modeling with limited data.
A BN model was developed for ICDA of a 50 km refined oil pipeline located in the western China. Internal corrosion probability of failure along the pipeline was assessed by quantifying the uncertainties using data provided by the pipeline operator. Because of the limited available data including lack of information on water content and other chemical compositions in the oil, the model predicts a low probability of failure for the entire 40-year service span simulated in the analysis. As a comparison, a what-if scenario analysis was carried out by assuming of a small amount of water in the oil. The results of which predict that the pipeline would fail in 15 years because of internal corrosion. In addition, a sensitivity analysis was employed to determine the most effective strategy to reduce the uncertainties. Results show that CO2, H2S, pH, and corrosion inhibition were the most important factors to define in order to improve the prediction accuracy of the internal corrosion model.
