|
Abstract
|
Pipelines are one of the most efficient and sustainable methods for transporting oil and gas, especially in offshore fields. Maintaining the integrity of pipelines, particularly against accidental threats, is a significant concern for owners of these critical assets and operational companies. Dropped object impact, as a common accidental pipeline threat, has the random nature of the collision in terms of the speed and mass of the dropped object, and the uncertainties arising from the pipe material, the strength of the concrete coating, and the pipeline corrosion conditions, that prioritize the examination and analysis of this significant phenomenon during the design stage. This article proposes a probabilistic model based on machine learning algorithms to calculate the risk associated with different levels of damage (based on the DNV-RP-F101 code) due to a dropped object impact onto the pipeline. A wide range of pipelines with their geometric and mechanical specifications, corrosion conditions, and various possible conditions of object impact with the pipeline have been considered to generalize the model. Machine learning algorithms, including Linear Regression, Decision Tree, Random Forest, K-nearest neighbors, Support Vector Machine, and Gradient Boosting, were investigated in this model, among which Random Forest has the highest accuracy. Investigating the susceptibility of the probability of different damage levels to pipeline characteristics is a basis in decision-making to take preventive measures to reduce the probability of pipeline damage occurrence at the design stage.
|