An assessment of causes and failure likelihood of cross-country pipelines under uncertainty using bayesian networks

•An integrated approach to dynamic pipeline failure likelihood analysis.•Incorporates subjective data and accommodates uncertainties using BN and AHP.•Identifies parameters that have the most impact on reducing pipeline loss of containment.•Nigeria's pipeline system used to show model application in situations where failure data is limited or unreliable. The increased incidents of pipeline failures and resultant consequences of fires, explosions and environmental pollution motivate stakeholders to find solutions in dealing with these emerging threats as part of process safety management. This is further compounded by the absence of reliable failure data, particularly in developing countries. To address such challenges, a Bayesian Network (BN) model has been developed. The aim of the model is to highlight the contributing failure factors to the identified pipeline hazards and their interrelationships. The BN approach is appropriate for this work because it accommodates data uncertainty, or the lack of data, and can integrate the expert's knowledge. The model is especially good at updating the results whenever new data becomes available. The proposed model has been applied to a case study focusing on estimating the failure probabilities of Nigeria's cross-country oil pipeline system - 2B as part of the pipeline risk assessment. The model takes into account multiple interactions between several failure parameters to reduce the risk of pipeline failure. Such parameters include human factors (e.g., third party intervention and operation damage), mechanical factors (e.g., corrosion and material defect) and natural hazards. The main focus of the research is the construction of a model that shows the influence of the multiple parameters and their interactions resulting in a pipeline leak or rupture. The model enables the pipeline stakeholders and operators to determine those parameters or interventions that have the most impact on the reduction in pipeline loss of containment as part of the risk management. The novelty of this work is the integration of both the objective and subjective data, and the explicit accommodation and treatment of the sparse and incomplete local data into the failure likelihood analysis. The model, therefore, provides the managers with dynamic information on how to prevent undesired outcomes as part of a safety management plan. The model analyses pipeline failure risks under uncertainty. However, it can also be used to focus on a sub-thre