Hybrid Parabolic Interpolation – Artificial Neural Network Method (HPI-ANNM) for long-term extreme response estimation of steel risers

Abstract

This paper presents a computer efficient approach to evaluate the multi-dimensional integral found in the evaluation of the long-term extreme response of marine structures. The proposed method is a hybrid combination of two numerical procedures. The first one consists of a parabolic interpolation scheme used to obtain the statistical parameters describing the short-term peaks probability distribution of the time-series responses, designated as PIM (Parabolic Interpolation Method), which reduces the total number of short-term structural analyses. The second one is an Artificial Neural Network-based surrogate model which is used to obtain long response time histories based on short finite element-based simulations. The approach is named as HPI-ANNM (Hybrid Parabolic Interpolation – Artificial Neural Network Method). The efficiency and accuracy of the proposed hybrid method is compared with the complete long-term integration method in the analysis of 100-yr characteristic values of cross-section utilization ratios of a Steel Catenary Riser (SCR) connected to a semi-submersible platform in deep water.