A Bayesian Network for Slope Geohazard Management of Buried Energy Pipelines

Abstract

Geohazards are the ground movement events that impose displacement demands on buried pipelines leading to excessive inelastic strains and possible loss of containment. Lack of data, as well as large uncertainties in the prediction of ground movement and inelastic pipe response are common challenges in geohazard management. A simple Bayesian network is presented in this study to demonstrate the integration of data from multiple sources, as well as prediction of the pipeline condition under a hypothetical ground movement scenario. The methodology used to obtain the network parameters and the challenges associated with the implementation are discussed. Geotechnical events resulting in ground movement, such as landslides and earthquakes, are termed ‘geohazards’ as they may lead to loss of containment in buried energy pipelines. Slope creep is the gradual soil movement at a slope due to changes in soil conditions, such as an increase of pore water pressure. Ground movements due to slope creep accumulate over the years and gradually increase the imposed displacements on buried pipelines, leading to excessive inelastic strains that can result in failure. As steel pipelines can sustain axial and bending strains beyond the elastic limit without immediate loss of containment, monitoring ground movements and pipe deformations at specified intervals provides an opportunity to reduce the probability of pipeline failure by identifying potentially critical strains and implementing slope remediation measures. Conventional approaches for pipeline structural integrity management are inadequate to address the uncertainties in assessing the probability of pipeline failure. Deterministic assessments are often conservative as uncertainties in the prediction of slope and pipe conditions are not considered explicitly (YoosefGhodsi et al. 2008). Both qualitative and semi-quantitative approaches focus heavily on the likelihood of slope movement using a combination of expert opinion and historical data of pipeline failures (PRCI 2009, Sen et al. 2018). Several approaches are available offering a framework to quantify failure frequencies as the product of conditional probabilities characterizing the sequence of occurrence of geotechnical events and only pipe size, wherein the required probabilities are often quantified based on expert opinion (Guthrie and Reid 2018, Baumgard et al. 2016, Ferris et al 2016, Porter et al. 2016). In quantitative assessments, a limit state function is defined as the tensile or compressive strain demand exceeding the strain capacity (also termed ‘strain limit’). Soil properties, slope parameters, and pipe parameters are characterized as random variables and used in empirical models to calculate the probability distributions of strain limits and pipe-soil interaction analysis to calculate the probability distributions of strain demand (Zhou 2012, Fraser and Koduru 2016, Koduru and Nessim 2018). However, strain demand can be difficult to predict, not only due to the large number of environmental factors controlling the amount of ground movement, but also due to the complexity of the pipeline’s response to the movement. Moreover, the collection of site-specific data 13th International Conference on Applications of Statistics and Probability in Civil Engineering, ICASP13 Seoul, South Korea, May 26-30, 2019 2 required for model inputs can require significant effort, particularly for slopes in remote locations. As part of pipeline structural integrity management against geohazards, the following multi-disciplinary data and domain expertise are used: isolated measurements of slope movement from slope inclinometers; estimates of terrain movement from satellite imagery or air photogrammetry, such as LiDAR, InSAR and Digital Elevation Models (Guthrie et al. 2018, Baumgard et al. 2014); pipeline strain gauge data (Dinovitzer et al. 2014); pipe curvature data from inline inspection (ILI); and results from detailed finite element modeling of the pipe-soil interaction (Fredj et al. 2015, Fredj et al. 2016). All of these tools and analysis methods come with varying levels of uncertainty and utilize data with significant temporal and spatial variability. Bayesian networks (BNs) offer a potential to address this diversity of information sources as this approach is capable of integrating data types from different sources and of different granularity, and has the abty to update the probability estimates based on new inspection data. 1. OBJECTIVE AND SCOPE The objective of the study described in this paper was to develop a BN for pipeline geohazards. The focus of the development was to demonstrate the integration of multiple data sources to predict the pipeline condition under a hypothetical ground movement scenario. Previous BN studies were limited to the assessment slope safety and did not include pipeline response modeling (Peng et al. 2014). The scope of the study is limited to ground movements that are primarily parallel to the pipeline axis, as shown in Figure 1. Slope creep along the longitudinal axis occurs most often at the pipeline water crossings (e.g. rivers and streams). As pipeline water crossings occur in all types of terrains – unlike ground movements perpendicular to the pipeline axis that are mostly limited to mountainous regions – this type of ground movement is of greater interest to pipeline integrity management. The paper provides a detailed discussion of BNs representing pipeline response to slow accumulated ground displacement at a specific slope and presents the results of performing Bayesian inference on the network to estimate probability of pipeline failure, and other pipe response conditions. The BN in the current study is modeled with cumulative ground displacements over a fixed time period instead of the incremental ground displacements. Figure 1: Ground Movement Direction Relative to the Longitudinal Axis of the Buried Pipeline At first, the development of network structure is explained, followed by the approaches used to develop the conditional probability tables needed to model the network. Next, a numerical example demonstrating the application of the proposed approach to a hypothetical pipeline is presented. The paper concludes with a discussion on the advantages and challenges associated with the development of BNs for slope movement and future work required to develop a comprehensive BN methodology to address geohazards. 13th International Conference on Applications of Statistics and Probability in Civil Engineering, ICASP13 Seoul, South Korea, May 26-30, 2019 3 2. NETWORK DEVELOPMENT 2.1. Event Nodes Figure 2 shows the sequence of data collection and analysis steps required to predict pipe condition in a moving slope. Data types used for inference through engineering judgment, structural mechanics and data analysis to estimate site and pipe parameters are listed in Table 1. In the development of a BN, it is of interest to identify the events related to pipe response, and monitoring information. In the present work, only those parameters related to inspection and monitoring are selected as event nodes. Although there are uncertainties due to inherent randomness in the physical properties and slope geometry, they are not modeled as explicit nodes in this study. Data collection related to these parameters is assumed to be complete with no expectation of updates due to new observations. Figure 2: Sequence of Steps to Assess Pipe Condition 2.2. Causal Links Figure 3 shows the full network diagram with events and links. As shown in the figure, event nodes are modelled to have discrete and finite states that are mutually exclusive and collectively exhaustive. The direction of influence between the event nodes is developed based on the mechanics of pipe-soil interaction under imposed differential ground movements and the sequence of pipe condition assessment events shown in Figure 2. When the slope geometry and buried pipeline elevation profile are known, pipe condition and failure modes depend primarily on cumulative ground displacement and sliding length. Table 1: Observed and Inferred Parameters. Parameter Direct Measurements Data for Inference Slope creep activity Field observations instrumentation Field observations, LiDAR, InSAR, ILI Slope failure mechanism Field observations Slope geometry, Geology, Field observations Ground displacement Field instrumentation Field instrumentation, LiDAR Sliding length Field observations LiDAR, Airphoto, InSAR, Slope Geometry Location of sliding block Field observations Slope inclinometers LiDAR, InSAR Soil strength Soil tests Surficial geology Pipeline alignment ILI As-built drawings, Right of way alignment Depth of burial Depth of cover survey Regulatory requirements Pipe material properties (grade, toughness) Material tests Pipe vintage, Design value Pipe dimensions (Size, wall thickness), Direct measurements Design values Internal pressure Readings at pumping stations Maximum allowable operating pressure Pipe strain Strain gauges, Pipe curvature from ILI Imposed loads, Pipe curvature from ILI 13th International Conference on Applications of Statistics and Probability in Civil Engineering, ICASP13 Seoul, South Korea, May 26-30, 2019 4 Table 2: Event Nodes.