Session: 14-02-02: Reliability and Risk in Energy Systems

Paper Number: 113465

113465 - Construction of a Strain-Based Bayesian Network for Assessing Pipeline Risk due to Ground Movement 

Ground movement events pose a significant threat to buried natural gas and oil pipelines and resulted in an estimated $388 million in damages from 2002 to 2022. Conventional stress-based design methods and failure theories struggle to accurately predict pipeline failure due to ground movement, and as such, the industry has increasingly adopted strain-based design and assessment (SBDA) methods to manage pipeline integrity in ground movement scenarios. However, utility companies still face many challenges in applying SBDA to ongoing pipeline integrity management: SBDA models require detailed metallurgical data which can be missing or expensive to collect, geotechnical data needed for pipe-soil interaction models is often missing, and SBDA methods are poorly standardized when compared to the current state-of-the-art.

Bayesian networks are probabilistic graphical models that document and quantify the causal relationships that exist within a system. Bayesian networks have become a useful tool in performing quantitative risk assessment for complex engineering systems for their ability to integrate data from multiple sources, perform probabilistic reasoning under uncertainty, and provide an understandable graphical representation of the variables and relationships within the model. These strengths make Bayesian networks well-suited to address the problems that utility companies are currently facing in applying SBDA methods.

This paper presents the development process of a Bayesian network structure for buried pipeline integrity management in a ground movement scenario. The Bayesian network has been partially parameterized, but its full parameterization and performance validation is ongoing work. To develop the Bayesian network structure, a hierarchical taxonomy was created to capture and organize the risk-influencing factors necessary for modeling SBDA. This taxonomy and previous SBDA models derived from multiple sources formed the foundation of the Bayesian network structure proposed in this work. Then, to parameterize the Bayesian network, several data were integrated, including satellite-based ground movement data, infield strain measurements, in-line inspection defect data, strain accumulation models, detailed metallurgical information on the behavior of the pipeline steel, and knowledge of the past and anticipated loading envelop of the pipeline. The parameterized model is capable of estimating, for a specified pipeline segment, its accumulated strain (strain demand), the maximum strain the pipeline can withstand without failing (strain capacity), and the probability of failure due to tensile leakage/rupture or compressive buckling. Furthermore, the analyst can use the model to perform probabilistic inference on the network, enabling scenario analysis for identifying suitable interventions to address pipeline risk.

The Bayesian network model proposed in this work presents a first-of-its-kind approach to integrate the multiple risk factors and data sources required to model SBDA and assess pipeline risk, identify the sources of uncertainty that may hinder an accurate SBDA, and provide a holistic tool for addressing pipeline risks stemming from ground movement events.

Presenting Author: Colin Schell University of Maryland, College Park

Presenting Author Biography: Colin is a third-year Ph.D. student in the Reliability Engineering program at the University of Maryland. Advised by Dr. Katrina Groth at the Systems Risk and Reliability Analysis lab (SyRRA), his research focuses on using causal models to better understand pipeline risks stemming from mechanical and natural hazard loading conditions, as well as third-party excavation. Colin also received his B.S. in mechanical engineering from the University of Maryland in 2020.

Authors:

Colin Schell University of Maryland, College Park
Ernest Lever GTI Energy
Katrina Groth University of Maryland, College Park