Predicting Residual Dent Parameters of Subsea Steel Pipelines Using Artificial Neural Networks

Abstract In the development of offshore oil fields, subsea pipelines are essential for various applications, but they are susceptible to accidents during operation, such as collisions with anchors, rocks, or heavy equipment, leading to mechanical damage like dents. This study aims to develop a robust methodology for predicting the residual depth, length, and width of dents, as well as evaluating deformation and stress concentration in damaged pipelines subjected to cyclic internal pressure loads, based on readily available field parameters. The proposed methodology integrates finite element analysis (FEA) with artificial neural networks (ANNs) to allow accurate and concise damage characterization that can be further used in a fatigue life assessment, for instance. A three-dimensional finite element model of a dented pipe was developed and validated against full-scale experimental tests conducted on API 5L Gr B steel pipe specimens. The calibrated model was used in a comprehensive parametric study, analyzing key geometric parameters of the pipe and dent across 240 FEA models that reflect standard industry geometries. This extensive dataset was then employed to train an ANN capable of predicting residual geometric parameters of the dent, the stress concentration factor (SCF), and the maximum equivalent plastic strain of dented pipe sections. The results indicate that ANNs can predict these parameters with accuracy comparable to FEA while offering substantial time savings. The immediate results provided by ANNs enhance their applicability across the entire pipeline system, including diagnostics and maintenance procedures, making them a valuable tool for optimizing pipeline integrity management.