Using Dynamic Bayesian Belief Network for analysing well decommissioning failures and long-term monitoring of decommissioned wells

There is increasing interest to consider dependent failures and human errors in the offshore industry. Permanently abandoned wells dot most of the subsea environment. The nature of a well plugging and abandonment (Well P&A) run is usually conducted in a manner such that the lowest-cost contractor is engaged to plug several wells tapping the same reservoir. Thus, this makes it an ideal case study for incorporating failures based on common causes. The heavy use of operators during a cementing job also provides the case for analysis of human error in such tasks. One proposed method to analyse the above-mentioned is the use of Bayesian Belief Networks (BBN) to achieve the following objectives (1) to capture better estimates of a well PA event by incorporating dependencies, and meet regulatory requirements by authorities; and (2) to use the same model to provide long term monitoring of a group of wells linked by common dependencies. This model has not only captured the dependencies of multiple variables, but also projected it in a dynamic manner to provide a risk profile for the next decade where well integrity failure is likely to happen. The sensitive analysis and backwards diagnostic analysis demonstrated that the results agree with a statistical study of 103 wells. This affirms that well failure probabilities are under-estimated if independent failure is assumed.