A BBN risk model of maintenance work on major process equipment on offshore petroleum installations

Operational safety is receiving more and more attention in the Norwegian offshore industry. Almost 2/3 of all hydrocarbon leaks on offshore installations in the period 2001-2005 according to the Risk level project by Petroleum Safety Authority in Norway, resulted from manual operations and interventions, as well as shut-down and start-up, confirming what is considered common knowledge; that incidents and accidents often are caused by failure of operational barriers. Investigations of major near accidents show that technical, human, operational, as well as organizational factors influence the leakages. In spite of these facts, quantitative risk analyses of offshore oil and gas production platforms have focused on technical safety systems. The intention with the Risk OMT (Risk Modelling-Integration of Organisational, Human and Technical factors) program has been to develop more representative models for calculation of leak frequencies as a function of the volume of manual operations and interventions. The Risk OMT program represents a further development of the work in the BORA and OTS projects. The emphasis is on a more comprehensive modeling of Risk Influencing Factors (RIFs) and how these affect the performance of operational barriers. The basic approach is the same, but the intention with the Risk OMT program has been to develop further the models used in BORA and OTS. In the Risk OMT project a generic risk model has been developed and is adapted to use for specific failure scenarios. The model considers the operational barriers in event trees and fault trees, as well as RIFs that determine the basic event probabilities in the fault trees. The generic risk model applies Bayesian belief networks in its modeling. The model has been evaluated through case studies and has been applied to evaluate the effect of different proposed strategies to reduce the leakage rates. This paper presents the model step by step from two viewpoints: the data and expert judgment needed, and how the model uses the provided data and judgments. Finally possible applications are discussed.