Accounting for Uncertainty in Predicting the Risk of Pipeline Failure. S. J. Arulanandam, D. J. Wilson, and S. E. Hrudey, Department of Mechanical Engineering, University of Alberta, Edmonton, AB T6G 2G8 and Department of Public Health Sciences, University of Alberta, Edmonton, AB T6G 2G3
The accidental release of toxic gases from pipeline failures is a rare event of significant interest to the pipeline industry. In this study we examine the tradeoffs between two approaches for estimating the risk of pipeline failure and we estimate the uncertainty involved in both approaches. One approach uses summary statistics from past failures to estimate the risk of future failure. In our study, analytical techniques are used to quantify the uncertainty in predicting risk based on aggregate pipeline data. The difficulty with this approach is that the resulting estimates of risk and uncertainty may be biased because the data have not been sorted to account for changing industry practices. For example, materials and maximum operating pressures of pipelines have changed over the past several decades. A second approach to estimating risk involves dividing the data into classes with similar characteristics and then making predictions based on these smaller data sets. As the data are subdivided to estimate the risk associated with particular characteristics, such as pipe diameter, corrosion factors, etc., the failures themselves become less frequent causing the uncertainty associated with such predictions to increase. Using existing data and pipeline statistics, we quantify the uncertainty associated with predictions of risk for specific types of pipelines or failure modes. The resulting estimates of uncertainty are then used to make comparisons between less precise risk estimates based on aggregate data sets and the higher uncertainty associated with estimating risks for particular failure classes.