When new tie-ins are made to existing infrastructure, corroded pipelines and risers may be subject to higher operating pressures than in their recent past. In order to help make informed decisions on whether such corroded components can be safely used at increased pressures, it is necessary to have a means of estimating their failure probability on introduction of the proposed higher operating pressure. A methodology has been developed to predict the probability of failure of a corroded pipeline or riser in the event of such an increase in operating pressure.
The application of the methodology to a 30-year old riser system containing several thousand corrosion clusters is described. The riser system comprised topside piping and a subsea tie-in spool as well as the vertical riser pipe. The analysis computed the probability of bursting if the pressure were to be elevated to a proposed new value. Using estimates of future corrosion rates, the analysis also estimated the remaining life of the system. The burst capacity model of DNV-RP-F101 was used, with allowance for the sizing accuracy of the inline inspection tool and for parameter and modeling uncertainties.
Two methods of computing the failure probability were explored:
First Order/Second Order Reliability Method (FORM/SORM) and
Monte Carlo Simulation (MCS). FORM and SORM are approximate methods, but very fast. MCS is an accurate method, but run times can be lengthy, particularly when the defect failure probabilities are low and when there are thousands of defects to analyze, as in the example case. As an exploratory exercise, FORM and SORM were tested against the MCS approach and found to have poor accuracy for this particular model. It was therefore decided to use MCS. The challenge in applying the MCS approach was then how to efficiently handle the analysis of thousands of corrosion clusters, each of which makes a small contribution to the overall failure probability.
The paper describes how the MCS method was efficiently implemented and how the finished program was used to investigate sensitivity of the analysis results to different modeling assumptions, such as whether variables are treated as independent or correlated.