Learning From Riser Analyses and Predicting Results With Artificial Neural Networks

Abstract

Subsea wellheads are subjected to fatigue loading from semi-submersible drilling vessels in harsh and relatively shallow waters like in the North Sea. Dynamic finite element riser simulations are run to ensure safe operations. These simulations calculate loads and fatigue damage on the subsea wellheads, and are run in large numbers, often with only small alterations in input. Building and running all these models for each new well are both time consuming and costly. By storing and structuring the results from such analyses, machine learning algorithms can be trained, and used to predict new results, without the need of running simulations for every new well. If there are insufficient simulation data available, data with inconsistent modeling, or little variety in the input of the data, it is also possible to use model builders to generate a sparse but sufficient set of simulation data to train the model. A trained model can predict simulation results instantly, and gains accuracy as more analysis data becomes available for training. The trained model is efficient for estimating fatigue status on oil fields with large numbers of subsea wells, since it is unnecessary with separate simulations for each well. It can also be used for early phase concept studies and as a QA tool for verifying results of other simulations.