Estimation of burst pressure of pipelines with interacting corrosion clusters based on machine learning models

Pipeline corrosion defects mostly appear in a colony such that they interact to reduce the failure pressure, which is not defined by features of a single corrosion defect. The huge amount of corrosion defects captured by in-line inspection tools including the variability of defect profile in pipelines and the dependence of the reliability assessment on such data pose significant research challenges in performance assurance. This highlights the need for computationally efficient modelling schemes to estimate the burst pressure of pipelines affected by both longitudinal and circumferential interacting corrosion defects. In the present paper, a novel approach is proposed for this purpose by combining supervised machine learning methods with 25 numerical models of corroded pipelines, validated with experimental results available from literature. Additionally, six improved composite defect shapes are proposed, resulting in 150 models to examine the non-linear behaviour of interacting corrosion defects by capturing the real the defect profiles captured by the In-line Inspection tools. The predicted failure pressures from the developed numerical models produced an absolute mean deviation of not exceeding 2.03% and 2.2% from the experimental burst pressure and the modified Mixed Type Interaction approach respectively, better than published results from the literature. Notably, the predicted failure pressures based on real pipeline data, infused with the generated artificial neural networks and non-linear regression models provide a total mean deviation of 3.1% and 7.3% respectively, thereby providing a path for effective maintenance planning.