Comparison Between Finite Element Model and an Artificial Neural Networks Procedure for Riser Analysis

Abstract

As exploitation activities moves into fields located in deep water, the industry has been addressing studies aiming at concepts of offshore systems that reduce the influence of environmental loads on risers. The Buoy Support Riser (BSR) system is one of these new proposed concepts. The BSR is composed by a subsurface tethered buoy, where flexible jumpers connect the Floating Production Unit (FPU) to the BSR and Steel Catenary Risers (SCRs). Due to its complexity and non-linearity, this offshore system requires a highly refined finite element model for dynamic analysis, which demands a high computational cost. In order to increase feasibility of the analysis it is proposed a low computational cost methodology based on Artificial Neural Networks (ANN). This work aims to develop a program to train an ANN to predict the jumpers’ dynamic tension from FPU motions without running the finite element model for every time step. In this way, the purpose is to find results as reliable as those achieved in a dynamic analysis with a finite element model. Statistical parameters will be used for this comparison.

Deep water drilling riser mechanical behavior analysis considering actual riser string configuration

“…Qi et al (2015) investigated the configuration of the drilling riser considering evacuation because of typhoon and recommended that the heavy wall thickness, such as 0.0254 m, and bare joint could be configured in the middle part of the riser system. de Aguiar et al (2015) proposed a low computational cost methodology based on artificial neural networks for riser analysis and argued that the results were as reliable as those achieved from finite element models. Connaire et al (2015) used quasi-rotations and the NewtoneRaphson method for riser analysis and showed that the approach provided advantages for subsea riser sections.…”