Sizing of a new build FPSO hull is an iterative optimization process, where the main dimensions of the FPSO are varied until the optimal size is found, within a defined input domain. A wide range of criteria should be evaluated per size iteration, and thus the sizing process may be a very time consuming task. In order to speed up this process, it may be beneficial to adopt an automated iterative algorithm for performing the sizing of new build FPSOs for the Concept and Front-End Engineering Design (FEED) phase.

The hull steel weight should be calculated for each size iteration, thus the hydrodynamic bending moment and shear force are needed. Obtaining the hydrodynamic sectional loads requires relatively time consuming calculations (compared to e.g. the required hydrostatic calculation), such as linear diffraction/radiation analysis and stochastic postprocessing. The required computational time of the automated sizing algorithm can be significantly reduced by calculating the hydrodynamic sectional loads by a simplified estimator, eliminating the need for e.g. diffraction/radiation analysis. Such an estimator may be obtained by the use of Artificial Neural Networks (ANNs).

This paper presents how to estimate the hydrodynamic vertical bending moment and shear force for FPSOs, within acceptable accuracy to be used for initial sizing, using ANNs. The estimators, i.e. the ANNs, take known inputs such as the main dimensions, draft and pitch radius of gyration of the FPSO. The ANNs are trained on a database containing linear diffraction/radiation analyses for a variety of FPSO main dimensions, hull shapes and loading conditions.

The database has been established by batch processing of the DNV-GL software HydroD [1], Wadam [2] and Postresp [3] through MATLAB [4]. This paper presents the methods used to obtain the results contained in the database, as well as training and performance of the ANNs.

This content is only available via PDF.
You do not currently have access to this content.