Abstract
Many engineers are inclined to not trust models; this is particularly true in the field of corrosion. Suspicion comes from modeled results which are inconsistent with field data. The difference between modeled results and the real world has three reasons. First, no model is accurate in all situations. Second, the input data used to run the models is never exact. And third, the operator's knowledge of the system is often missing. In order to increase confidence and reduce the gap between modeled results and field data it is necessary to address all three sources of uncertainties.
A solution is proposed: (1) never trust one model only, run multiple models, (2) run the models multiple times for all possible input parameters (Monte Carlo method) and (3) combine the output of different models as well as the operator's knowledge in a graphical interface (Bayesian network). The results of the multiple simulations are not numbers, but a function of all the possible outcomes predicted by the models.
This paper presents a Bayesian network created to assess the probability of pipeline failure due to internal corrosion. The model created quantifies the likelihood (and uncertainty) of pipeline failure as well as all causative factors. For a pipeline operator it is necessary to reduce both the likelihood of failure and the uncertainty that comes with it. The uncertainty is reduced by gathering more data and the likelihood is reduced by mitigation. Therefore, results of the Bayesian networks can be used both to prioritize inspections (reduce uncertainty) and to prioritize mitigation (reduce likelihood of failure). The Bayesin network model was implemented and successfully verified on a crude oil export pipeline operated by an oil company in the Middle East.
