Towards a causal model from pipeline incident data analysis

Abstract

Pipeline network facilitates transportation of hazardous materials over long distances and is considered a relatively safer mode of transportation. However, a significant number of incidents occurred over the last few years that has increased the need for development of an accurate predictive model for incidents. Pipeline incident databases have been collecting a wide range of data regarding the incidents including the background operating conditions, the causes contributed to the incident, and the severity of consequences. Most academic studies analyzing these databases have developed predictive failure models focusing only on a narrow range of factors that played a role behind certain categories of incidents. They fail to provide a holistic understanding of how interaction of multiple factors contributing to an incident or how their combined effect eventually leads to failure. The amount of data collected in the databases are however huge. For example, the current reporting format of Pipeline and Hazardous Materials Safety Administration (PHMSA) collects over six hundred datapoints from each incident. Given the large amount of information available and current technologies, it is now possible to look into ways to understand the holistic effect of all contributing factors. The current study looks into how the databases can be utilized to derive such a complete understanding. The study compares the databases of US PHMSA, Canada National Energy Board (NEB), and European Gas Pipeline Incident Data Group (EGIG) to examine the frameworks used to classify the causal factors of the incident data and shows that the background factors, underlying factors and the causal factors identified in the various databases are inter-related and show varying degrees of dependency. This curves the way forward to development of a causal model that will help identify the important factors that needs to be addressed and a prediction of future failure given existing conditions of a pipeline system.