Bayesian network model for predicting probability of third-party damage to underground pipelines and learning model parameters from incomplete datasets

Abstract

Damage caused by third-party excavation is one of the leading threats to the structural integrity of underground energy pipelines. Based on fault tree models reported in the literature, the present study develops a Bayesian network (BN) model to estimate the probability of a given pipeline being hit by third-party excavations by taking into account common protective and preventative measures. The Expectation-Maximization (EM) algorithm in the context of the parameters learning is employed to learn the parameters of the BN model from datasets that consist of individual cases of third-party activities but with missing information. The effectiveness of the parameter learning for the developed Bayesian network is demonstrated by a numerical example involving simulated datasets of third-party activities and a case study using real-world datasets obtained from a major pipeline operator in Canada. The BN model and EM-based parameter learning proposed in this study allow pipeline operators to estimate the probability of hit by efficiently taking into account historical third-party excavation records in an objective, efficient manner.