Artificial Neural Networks for Reducing Computational Effort in Active Truncated Model Testing of Mooring Lines

Abstract

As oil and gas exploration moves to deeper waters the need for methods to conduct reliable model experiments increases. It is difficult obtain useful data by putting a scaled model of an entire mooring line systems into an ocean basin test facility. A way to conduct more realistic experiments is by active truncated models. In these models only the very top part of the system is represented by a physical model whereas the behavior of the part below the truncation is calculated by numerical models and accounted for in the physical model by active actuators applying relevant forces to the physical model. Hence, in principal it is possible to achieve reliable experimental data for much larger water depths than what the actual depth of the test basin would suggest. However, since the computations must be faster than real time, as the numerical simulations and the physical experiment run simultaneously, this method is very demanding in terms of numerical efficiency and computational power. Therefore, this method has not yet proved to be feasible. It has recently been shown how a hybridmethod combining classical numerical models and artificial neural networks (ANN) can provide a dramatic reduction in computational effort when performing time domain simulation of mooring lines. The hybrid method uses a classical numerical model to generate simulation data, which are then subsequently used to train the ANN. After successful training the ANN is able to take over the simulation at a speed two orders of magnitude faster than conventional numerical methods. The AAN ability to learn and predict the nonlinear relation between a given input and the corresponding output makes the hybrid method tailor made for the active actuators used in the truncated experiments. All the ANN training can be done prior to the experiment and with a properly trained ANN it is no problem to obtain accurate simulations much faster than real time — without any need for large computational capacity. The present study demonstrates how this hybrid method can be applied to the active truncated experiments yielding a system where the demand for numerical efficiency and computational power is no longer an issue.