A reliable and robust means of continuously monitoring the integrity of mooring systems for FPSOs and FPUs is required in the offshore industry. The consequences and costs associated with a seriously degraded mooring system, resulting in an undetected failure, can result in a riser system failure, extended production downtime and environmental damage from hydrocarbon spillage.
This paper presents a novel method to continuously track and predict the performance of a moored floating body under the full spectrum of metocean conditions, and uses this capability to rapidly detect a change in the system’s performance (eg a single mooring line failure or a loss of station event.) These events are often not identified offshore when they occur, as a small change in displacement from a nominal mooring center is difficult to detect without a reference point, especially when the facility is subjected to a sustained mean offsets due to the prevailing metocean conditions.
Conventional methods of predicting the motions of a floating body usually involve simulating wind, wave, current and sea level in a hydrodynamics packages. These packages are cumbersome to use in the real time assessment of FPSO response due to live measurements of metocean conditions, as these packages are typically complex to set-up a model and are subjected to lengthy computational run time. An alternative is a “black-box” system; one that uses machine learning algorithms to predict the motions of the floating body by being trained with the measured metocean inputs and the FPSO response. Several algorithms were investigated for this purpose, and two leading candidates investigated further; Kriging and Neural Networks. Both methods were used to as they perform well when used predict complex multivariate systems.
To train both these learning algorithms metocean and GPS data from an FPSO was available. This dataset contained two distinct parts; a phase where the mooring system was degraded and a phase when the mooring configuration changed due to the repair of some of the mooring legs. This mooring configuration change provided a distinct change in the systems, which was taken to be analogous to a mooring line failure event. It would be expected that a change in vessel response would occur due to a change in the mooring system configuration, especially if the system responds due to second order wave forcing.
Both learning methods were trained on a subset of the collected data and the accuracy of each validated by a second distinct subset of the data, with the same mooring configuration. Both models were able to accurately predict the motions of the FPSO in the time and frequency domain. Models that were trained on the original data were then passed data for the changed mooring configuration. Both methods were able to successfully identify the mooring system change by accurately predicting the motion of the FPSOs for the mooring configuration they were trained on, and did not predict a similar response for the dataset with the different mooring configuration.
The methods investigated were found to reliably predict the motions of the FPSO as the learning algorithms provided a means of detecting changes in state of the mooring system. These predictions using the studied learning algorithms provided a more robust means of continuously monitoring the integrity of mooring systems for FPSOs and FPUs than subsea instrumentation systems or resorting to real-time computing using classical hydrodynamic analyses.