Simulated Full Lifetime Response Data of a Turret-Moored FPSO for Training AI Using HPC

Abstract

Typically, limited operational time and data are available to adequately train artificial intelligence (AI) models for new field developments. Simulation data has the potential to train AI, but it must be sufficiently accurate, reliable and comprehensive. The main objective of this research is to generate and customize response data of a turret-moored Floating Production Storage and Offloading (FPSO) vessel for AI training. Direct time-domain simulation, covering the entire service life (e.g. 100,000 simulations, each representing 3 hours, for 35 years) of a floating platform, has become practical using High Performance Computing (HPC). In this study, 21-year hindcast Gulf of Mexico environmental data with 3-hour intervals for the Gulf of Mexico are simulated, and the responses of a turret-moored FPSO are analyzed using a fully coupled time domain analysis method. The FPSO and turret are modeled as independent bodies and connected through a bearing connection model, which allows rotation of the FPSO with respect to its mooring system. The numerical model, which has been validated through model tests, is utilized for simulating the entire service life responses of a turret-moored FPSO. The results provide 61,360 cases of 3-hour time series data with 6 degrees of freedom (DOF) responses for both the turret and FPSO, mooring tensions, and interaction loads between the turret and FPSO. Significantly, the large yaw response, which is a unique characteristic of a turret-moored FPSO, is accurately captured. Moreover, the simulated data set is sufficiently large for AI training, and the real time predictions of a turret-moored FPSO are discussed.