River Crossing Reliability Screening Analysis Using a Bayesian Framework

Abstract Pipelines cross tens of thousands of river and stream crossings throughout North America. Scour of the riverbed material at these crossing locations can cause buried pipelines to become free-spanning and susceptible to vortex induced vibration fatigue or overstress failure. Current reliability assessment approaches require extensive (20+) site specific inputs, which correspondingly require detailed desktop review and/or field visits. Obtaining the data required to conduct these types of detailed assessments at every stream crossing is often neither feasible nor practicable for entire pipeline networks. In this paper, the authors present a systemwide screening analysis approach to estimate the probability of failure of pipelines at river and stream crossings, requiring only 3 input variables but expandable to include more input variables. In this way, crossings that exceed a particular target can be prioritized for more detailed assessment. Moreover, a key strength of the model is modularity: the number of inputs can vary depending on the known characteristics of the crossing being assessed, which is beneficial at small streams where detailed inspections have not been conducted. Conversely, results are enhanced using additional inputs at crossings where more information is known. The model uses historical industry failure observations as its basis and employs a Bayesian updating approach. Conditions that led to previous failures are compared to characteristics at crossings of interest to evaluate their probability of failure. A benefit of the approach is, unlike index-based models which can rely on subject matter expert judgement for index weightings, the Bayesian approach uses known data from historical failures as a fundamental basis. Distributions are fit to the historical failure data (e.g. bankfull width, measured depth of cover prior to failure, pipeline diameter) and the characteristics of the fitted distributions directly influence the response of the model, which eliminates the need for expert judgement in setting index factor weights. In consideration that only minimal input data is required, the model was applied to all 5,000+ river crossings across Enbridge’s liquids pipeline mainline system. The screening model results were compared to the currently available detailed models results and showed reasonable agreement. It was also shown that approximately 90% of known stream crossings did not warrant detailed analysis under their current conditions.