Risk assessment of decommissioning options for offshore facilities

• Main Author: Fam, Mei Ling
• Other Authors: Ong Lin Seng
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Systems engineering
• Online Access: https://hdl.handle.net/10356/143055
• Institution: Nanyang Technological University

Decommissioning is gaining traction globally. Large project management is required to plan and execute an offshore decommissioning operation. There is also increasing interest to adopt risk assessment techniques from other industries for the offshore industry, such as in the area of dependent failure or the role of human error. Finally, piece-wise, segmental risk assessment is based on traditional risk assessment methods, without consideration of long-term effects of a decommissioned structure. This thesis proposes a risk model that consists of three parts: (i) decommissioning options analysis (ii) selected decommissioning activity risk analysis and (iii) long-term monitoring of a decommissioning solution. A case study to a well plugging and abandonment is carried out as that constitutes the largest proportion of decommissioning costs and that well leaks are underestimated. The thesis has three research objectives. The first objective is to develop an expert judgement process that can consider expert uncertainty and inherent variability in the situation that warrants an expert judgement. Decommissioning projects rely a lot on expert judgement to choose the most apt decommissioning option. The second objective is to model dependencies in terms of time dependent failure (Event 2 failure given Event 1 failure) or Common Cause Failures (CCF) in greater details (material, environment, design). This objective is meant to provide a higher level of detail amongst CCF groups/causes to give insight to relationships between events that is not covered in existing simple CCF-ratio models, where the CCF probability is a ratio of the failure probability of the basic event. The third objective is to develop long-term monitoring to consider accumulative fatigue in the annular fit and the casing strength. This allows the projection of the well barrier failure rate to a decade, which is when a well would most likely fail, if it were to fail. A risk modelling method that combines and adapts methods from different and/or similar industries have been proposed. The method comprises of the following modelling features in a Bayesian Belief Network (BBN): (i) multiple states in a node (ii) dependency analysis (iii) representation of uncertainties (iv) detailed HRA (v) continuous variables and (vi) temporal variables. Expert judgement is a crucial part of the risk modelling process in the decommissioning field. Such judgement is utilised in options analysis, comparative assessments, human reliability analysis and discrete Markov chain time slices. Expert judgement is the primary input to the BBN. This is a delicate process as it should be comprehensive enough to avoid different types of biases, yet not be overly unpractical. The method adapted from the combination of the linear interpolation method and the Bayesian aggregation method can demonstrate dominance of one factor over the other, or represent different uncertainties in relationships. The modelling is also conducted in two specialised BBN software called GeNIe and Agenarisk. The model is able to carry out a weighted sum of the probability across all possible intervals and with key values of indication of uncertainty - such as the median value, or the 5th or 95th percentile. The method also utilised linear interpolation to reduce the elicitation burden. The results demonstrate that the interpolation method still manages to capture the dependencies and that the obtained HEP values agrees with other HRA-BBN models. The model also has a dynamic component that can provide insights on long-term well failures. Since fatigue in cement is an accumulative process, dynamic BBN (discrete-time Markov model) modelling has been utilised to investigate the performance of barrier over time, in this case, the annular fit and the casing strength. The model incorporated two common cause failures: (i) changes in wellbore pressure and the (ii) change in annulus pressure acting on top of the cement, which ultimately affects the fit of the cement with respect to the well bore. The results of the sensitivity analysis and backwards diagnostic analysis agree with a statistical study of 103 wells. Decommissioned structures are meant to last in perpetuity but the proposed solutions usually do not last in perpetuity. The proposed methodology of dynamic Bayesian Belief Networks (1) captured better estimates of a well PA event by incorporating dependencies, and met regulatory requirements by authorities; and (2) utilised the same model to provide long term monitoring of a group of wells linked by common dependencies. This addressed the challenges of underestimating failure probability in plugged and abandoned wells both in the short and long term.