Deep-learning approach based on multi-data fusion for damage recognition of marine platforms under complex loads

Abstract

This paper presents an innovative deep-learning network underpinned by advanced data-fusion techniques developed explicitly for detecting structural damage in marine platforms. This study commences with a rigorous validation of the proposed methodology employing detailed experimental procedures and subsequently applies this method to identify damage under specific random load conditions utilizing state-of-the-art finite element analysis software. This entails the meticulous construction of an extensive database mirroring the real-life conditions encountered by a deep-water jacket platform in the South China Sea. By conducting thorough simulations based on the characteristic environmental loads of this marine area, the study succeeds in closely replicating real-world conditions. The developed damage-identification model, which is specifically adapted to the distinctive requirements of non-Gaussian white-noise loads, exhibits notable accuracy. The experimental outcomes derived from the application of the proposed deep-learning algorithm for data fusion indicate an impressive accuracy rate of 92.01%. In simulated scenarios involving complexly loaded jacket platforms, the model attains an accuracy of 88.25%, highlighting the robustness of the method and its wide applicability in the field of marine structural health monitoring.